Some guidelines for Genetic Algorithm implementation in MINLP Batch Plant Design problems

Abstract

In the last decades, a novel class of optimisation techniques, namely metaheuristics, has been developed and essentially devoted to the solution of highly combinatorial discrete problems. The improvements provided by these methods were however extended to the continuous or mixed-integer optimisation area. This paper addresses the problem of adapting a Genetic Algorithm (GA) to a Mixed Integer Non Linear Programming (MINLP) problem. The support of the work is optimal batch plant design, which is of great interest in the framework of Process Engineering. This study deals with the two main issues for GAs, i.e. the treatment of continuous variables by specific encoding and the efficiently constraints handling in GA. Various techniques are tested for both topics and numerical results show that the use of a mixed real-discrete encoding and a specific domination based tournament method constitutes the most appropriate approach.