Simulação e técnicas da computação evolucionária aplicadas a problemas de programação linear inteira mista

Abstract

Presently, companies live a reality of rapid economic transformations generated by globalization. The growth of the products and services international trade, the constant exchange of information and the cultural interchange challenge administrators to define new paths for their companies. This dynamics and the increasing competitiveness demand new knowledge and abilities from professionals. In this way, new technologies are researched in order to improve operational efficiency. The Brazilian oil industry in particular has invested in applied research, as well as on development and technological qualification to keep its competitiveness in the international market. Many are the problems that must still be studied in this production sector. Among these, and due their importance, the problems of products storage and transference can be pointed out. This work approaches a scheduling problem that involves diesel oil storage and distribution in an oil refinery. The Mixed Integer Linear Programming (MILP) techniques with representation in the discrete and continuous time were used. The models that were developed were solved by the LINGO 8.0 software, using the branch and bound algorithm. However, due to their combinatorial nature, the expended computational time used for thesolution was excessive. Thus, four new methodologies were developed: Hybrid Steady State Genetic Algorithm (HSSGA) and Transgenetic ProtoG Algorithm, both integrated to Linear Programming (LP), for the representation of discrete time; simulation with optimization using the Genetic Algorithm (GA) and simulation with optimization using the Transgenetic ProtoG Algorithm, for the representation of continuous time. The results obtained through several tests with these new methodologies have shown that they can reach good results in an acceptable computational time. The two techniques for the representation of discrete time have shown satisfactory performance in terms of quality of solution and computational time. Among these, the methodology that uses the Transgenetic ProtoG Algorithm showed the best results. Also, the simulator with optimization using GA and the one that used the Transgenetic ProtoG Algorithm for the representation of continuous time were adequate to substitute the resolution through PLIM, because they reach solutions with a reduced computational time when compared with the time used for the solution with branch and bound.