Computers and Chemical Engineering 25 (2001) 257 – 266 Evolutionary algorithms approach to the solution of mixed integer non-linear programming problems Lino Costa, Pedro Oliveira * Department of Production and Systems Engineering, Uniersity of Minho, 4710 Braga, Portugal Received 7 January 2000; received in revised form 25 October 2000; accepted 25 October 2000 Abstract The global optimization of mixed integer non-linear problems (MINLP), constitutes a major area of research in many engineering applications. In this work, a comparison is made between an algorithm based on Simulated Annealing (M-SIMPSA) and two Evolutionary Algorithms: Genetic Algorithms (GAs) and Evolution Strategies (ESs). Results concerning the handling of constraints, through penalty functions, with and without penalty parameter setting, are also reported. Evolutionary Algorithms seem a valid approach to the optimization of non-linear problems. Evolution Strategies emerge as the best algorithm in most of the problems studied. © 2001 Elsevier Science Ltd. All rights reserved. Keywords: Evolutionary algorithms; Genetic algorithms; Evolution strategies; Mixed integer non-linear programming www.elsevier.com/locate/compchemeng 1. Introduction Real world problems can, in general, be formulated as mixed integer non-linear programming problems (MINLP). These problems, due to their combinatorial nature, are considered difficult problems. Gradient op- timization techniques have only been able to tackle special formulations, where continuity or convexity had to be imposed, or by exploiting special mathematical structures. Approaches based on stochastic algorithms have been used. These approaches, also know as adap- tive random search, have successfully tackle MINLP, mostly in the area of chemical engineering (Reklaitis, Ravindran, & Ragsdell, 1983; Salcedo, 1992; Banga & Seider, 1996). In recent years, a vast amount of work has been published on applications of evolutionary algorithms (Genetic Algorithms, Evolution Strategies, Simulated Annealing, etc.) to the solution of MINLP in many engineering applications. These algorithms are distinct from conventional algorithms since, in general, only the information regarding the objective function is required. Moreover, they start from a pool of points that evolves over time, possibly in the direction of the optimum. It should be stressed that the objective of this on going research is not to find the ‘best’ algorithm for all problem instances, but to compare different al- gorithms, in order to find out classes of problems which may be more suitable for certain algorithms than others. The general formulation of the problem is as follows: min f (x ) subject to h (x ) =0 (1) g (x ) 0 x j integer, j I x X ={x x R n , x l x x u } where h R m , g R p . In this work, 7 test problems, proposed by indepen- dent authors, are studied using Genetic Algorithms (GAs) and Evolution Strategies (ESs). These problems arise from the area of chemical engineering, and repre- sent difficult non-convex optimization problems, with continuous and discrete variables. Comparisons are made with an M-SIMPSA algorithm (Cardoso, Sal- cedo, & Feyo de Azevedo, 1996a,b; Cardoso, Salcedo, * Corresponding author. Fax: +351-53-253604741. E-mail address: pno@dps.uminho.pt (P. Oliveira). 0098-1354/01/$ - see front matter © 2001 Elsevier Science Ltd. All rights reserved. PII: S0098-1354(00)00653-0