Metaheurística e sistema de suporte à decisão no gerenciamento de recursos florestais

Abstract

The objectives of this work were: to evaluate the contribution of models, data, technologies and human resources in the process of decision making for forest management; to develop and present a decision support system with new approaches to even-age forest management; and to develop and test the metaheuristic Genetic Algorithm (GA), Tabu Search (TS) and Simulated Annealing (SA) to solve problems of forest planning with intiger constraints. To accomplish these objectives, this study was divided in six chapters. Chapter 1 comprises a review on the use of models, data, technologies and human resources to make decision in forest management. The study reveals the need of considering the integration of those resources, showing still the major drawbacks and misunderstandings in their use by forest companies. Chapter 2 presents SisFlor, which is a flexible Decision Support System (DSS) with a friendly interface to the user that allows to formulate and solve via Linear Programming (LP) and Integer Programming (IP) models important issues of even-age forest management. The limitations of the exact algorithm branch and bound to solve IP models, formulated through SisFlor DSS, is the main reason to develop the following chapters. Chapter 3 presents a review on the heuristics, the most promising approaches for solving IP problems. Among the heuristic approaches, the metaheuristic GA, TS and SA were selected and discussed on the basis of their basic principles and of some applications to forest management. In chapters 4, 5 and 6 the metaheuristic GA, TS and SA were formulated and implemented into a computational code. To test these metaheuristics, five problems were selected containing between 12 and 423 decision variables, with constraints of singularity and minimum and maximum production. All the problems had the objective of maximizing the net present value. Problem one was used for illustrating a race of the concerning metaheuristics. The other problems were used to evaluate the effects of several parameters of those metaheuristics, that were evaluated according to a efficacy measure given by the relation between the best value of the objective function obtained by the metaheuristic and the mathematical optimum obtained by the exact algorithm branch and bound. The parameters of the evaluated metaheuristics were compared by the L&O test and the analysis by descriptive statistics. The best parameter values selected for the concerning metaheuristics gave average efficacy of 95.97% for TS, 95.36% for SA and 94.28% for GA. In relation to the maximum efficacy values obtained, SA presented larger value for one of the problems (100%), followed by TS (98.84%) and GA (98.48%). The largest coefficient of variation was obtained by SA (3.18%), followed by TS (2.48%) and AG (2.08%). The efficiency, which was measured by the processing time for obtaining the solution through TS, GA and SA, was about two, five and ten times superior to the algorithm branch and bound respectively. Those metaheuristics were shown as attractive approaches for the solution of important combinatorial problems within the forest management context, problems of difficult or even impossible solution by the current exact algorithms.