Carlos A. Mendez rescheduling (constructive / improvement) multistage batch scheduling

http://cepac.cheme.cmu.edu/pasi2008/slides/mendez/index.html

Resource-constrained scheduling.gms

* An MILP based technique for a resource-constrained scheduling problem

BINARY VARIABLES
                    X(o,o0)    sequencing variable of product p and p0
                    A(o,u)     allocation of product p to unit u
                    B(o,m)     allocation of product p to operator m

* Activate preordering by slack time.
*$ontext
parameter slt(o) ;
slt(o) = dd(o) - smin(u $pt(o,u),pt(o,u)) ;
X.fx(o,o0) $(ord(o)<ord(o0) and slt(o)>slt(o0)) = 0 ;
X.fx(o,o0) $(ord(o)<ord(o0) and slt(o)<slt(o0)) = 1 ;
*$offtext



large scheduling problem.gms
*solve directly

* how to get this?
TABLE Allocation(P,S)  Unit allocated in each stage in the initial schedule

BINARY VARIABLES
                    X(p,p0,s)  sequencing variable of product p and p1 in stage s
                    X1(p,p0)   sequencing variable of product p and p1 in all stages
                    A(p,m)     allocation of product p to unit m ;

solve schedule minimizing MK using MIP;

* UPDATE POSITION FOR ALL BATCHES
loop(s,
    loop(m $(stagemach(s,m) > 0) ,
         loop(pr1 $(prodmach(pr1,m) > 0 and A.L(pr1,m) = 1 ),
                             pos = 1 ;
                             loop(pr2 $(prodmach(pr2,m) > 0 and FT.L(pr2,s) < FT.L(pr1,s) and A.L(pr2,m) = 1 ),
* pr2 is before pr1
                                  pos = pos + 1 ;
                                 );
                             Position(pr1,s) = pos ;
              );
        );
   );



improvement method.gms
** constructive and improvement

** 0 model:
** Define processing sequence for an already scheduled order and a released order
** Variable sequence and allocation for the released order

** Define processing sequence for a pair of already scheduled orders
** Fixed sequence and allocation for both orders

* 1 Constructive Schedule (insert one by one )
 Loop(pr3,
* Set the released table to zero
          loop(pr1,
                rel(pr1) = 0 ;
               ) ;

* (insert one by one )
          rel(pr3) = 1 ;
          pos1 = ord(pr3) ;

          solve pharma minimizing TMK using MIP;

* UPDATE ALLOCATION UP TO RELEASED BATCH
* UPDATE POSITION UP TO RELEASED BATCH

* 2 improvement stage
* Improvement optimization-based technique for the scheduling of a multistage Pharmaceutical process

OPTION reslim = 30 ;

* BEST CURRENT OBJECTIVE VALUE (starting point)
SCALAR BEST  ;

BEST = TMK.L ;
*
* Maximum number of iterations without changes
SCALAR MAXIT   / 30 / ;

***************************************************************
 WHILE (ITER < MAXIT ,

* Set the released table to zero
          loop(pr1,
                rel(pr1) = 0 ;
               ) ;
* Define the batch to be released
          BATCH = BATCH + 1 ;
          IF ( BATCH > CARD(p),
               BATCH = 1 ;
             );

* Release batch
           loop(pr1 $(BATCH = ord(pr1)),
                  rel(pr1) = 1 ;
               ) ;

* INCREASE THE NUMBER OF ITERATIONS BY 1
          ITER = ITER + 1 ;

          solve pharma minimizing MK using MIP;
          reschedtime = reschedtime + pharma.resusd ;

          IF ( (pharma.modelstat = 1 or pharma.modelstat = 2) and
                 MK.L < BEST ,
                 pharma.cutoff = MK.L ;
                 put batch:<2:0 , MK.L:>9:3 / ;
                 BEST = MK.L ;
                 ITER = 1 ;
* UPDATE ALLOCATION FOR RELEASED BATCHES
* UPDATE POSITION FOR ALL BATCHES

* Generate the best schedule that was found for output report