To meet fresh water requirement as well as to tackle wastewater generation are serious issues in process industries in the present scenario. To deal with water network problems, various models and computer aided tools have been used. GAMS software is used for this purpose since many years. All kind of water network models including LP, NLP, MILP and MINLP are extensively solved by this tool. However, it appears that no specific work is available in literature to observe the behaviour of different solvers on industrial case studies. The present study addresses this gap through behaviour analysis of LP and MILP solvers. Four case studies are considered from open literature, which include water networks for single contaminant, multi-contaminant and process as well as treatment units. LP models of all cases are solved through three different solvers i.e. CPLEX, CONOPT and MINOS. Further, piping cost is incorporated into LP models and thus, converted these to MILP, which are solved using three solvers such as BDMLP, CBC and SCIP. Results obtained through six solvers are compared at BECOTA scale. MINOS performs better for LP models as compared to CPLEX- a traditionally used LP solver. However, CBC is better amongst other MILP solvers considered. It is observed that selection of a particular solver may significantly influence the water network design as well as plant economy. Hence, selection of solvers should be carried out carefully.
Read full abstract