TY - JOUR
T1 - Long range planning in the process industries
T2 - A projection approach
AU - Liu, Ming Long
AU - Sahinidis, Nikolaos V.
N1 - Funding Information:
Acknowledgement--Partial financial support for this work has been received from the Ministry of Education, Taipei, Taiwan, R.O.C., the National Chengchi University, Taipei, Taiwan, R.O.C. and the Department of Mechanical and Industrial Engineering of the University of Illinois. Thanks are due to Professors Egon Balas and John Hooker for useful suggestions at an early stage of this work. The authors are also indebted to anonymous referees whose numerous suggestions improved the quality of this paper considerably.
PY - 1996/3
Y1 - 1996/3
N2 - The problem of selecting processes and capacity expansion policies for a chemical complex can be formulated as a multiperiod, mixed-integer linear program (MILP). This MILP can be reformulated by exploiting lot sizing substructures. The reformulation produces a tight linear programming relaxation but also introduces a large number of variables and constraints. To avoid unnecessary additional variables and constraints, a polyhedral approach is developed. The reformulation variables are projected out giving rise to a larger constraint system. The new model is solved using a strong cutting plane approach. The separation problem is solved exactly in polynomial time. Computational experiments are performed to compare the conventional MILP formulation with the nonconventional formulations after variable disaggregation and projection. It is found that only a few of the constraints of the projection model suffice to substantially reduce the relaxation gap of the conventional model. This property leads to more robust and faster solution algorithms for large scale problems. Computational results are presented for planning problems with up to 38 processes, 25 products and 8 time periods. The corresponding MILPs included up to 300 binary variables and a few thousand continuous variables and constraints.
AB - The problem of selecting processes and capacity expansion policies for a chemical complex can be formulated as a multiperiod, mixed-integer linear program (MILP). This MILP can be reformulated by exploiting lot sizing substructures. The reformulation produces a tight linear programming relaxation but also introduces a large number of variables and constraints. To avoid unnecessary additional variables and constraints, a polyhedral approach is developed. The reformulation variables are projected out giving rise to a larger constraint system. The new model is solved using a strong cutting plane approach. The separation problem is solved exactly in polynomial time. Computational experiments are performed to compare the conventional MILP formulation with the nonconventional formulations after variable disaggregation and projection. It is found that only a few of the constraints of the projection model suffice to substantially reduce the relaxation gap of the conventional model. This property leads to more robust and faster solution algorithms for large scale problems. Computational results are presented for planning problems with up to 38 processes, 25 products and 8 time periods. The corresponding MILPs included up to 300 binary variables and a few thousand continuous variables and constraints.
UR - http://www.scopus.com/inward/record.url?scp=0030107889&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0030107889&partnerID=8YFLogxK
U2 - 10.1016/0305-0548(95)00028-3
DO - 10.1016/0305-0548(95)00028-3
M3 - Article
AN - SCOPUS:0030107889
SN - 0305-0548
VL - 23
SP - 237
EP - 253
JO - Computers and Operations Research
JF - Computers and Operations Research
IS - 3
ER -