Estudo comparativo de metaheurísticas aplicadas ao problema de planejamento da produção florestal

Abstract

This work presents a comparative analysis of metaheuristic methods applied to forest production planning, aiming to maximize the Net Present Value (NPV) on a base of 120 forest management units and 81 management alternatives over 16 years. To tackle the high combinatorial complexity and operational constraints of the sector, algorithms of different natures were selected, such as the Genetic Algorithm (GA), Ant Colony Optimization (ACO), Simulated Annealing (SA), Tabu Search (TS), and Variable Neighborhood Search (VNS). To validate the effectiveness of these approaches, tests were conducted in three scenarios with different levels of minimum demand (Dmin) and maximum demand (Dmax) per year. The results indicated that all approaches were able to generate feasible solutions in Scenario 1 (Dmin = 140.000 m3 and Dmax = 160.000 m3) and Scenario 2 (Dmin = 140.000 m3 and Dmax = 150.000 m3), however there are no valid solutions in Scenario 3 (Dmin = 150.000 m3, Dmax = 160.000 m3). In addition, the algorithms showed significant differences in terms of NPV means, stability, and sensitivity to constraints. In general, the ACO and SA methods presented the highest average NPVs in Scenarios 1 and 3, while the GA stood out for recurrently presenting the highest number of runs with valid solutions, evidencing greater robustness in the face of the problem constraints. However, in Scenario 2 all algorithms presented similar averages, with no statistically significant differences. Thus, the study reinforces the potential of metaheuristics as effective tools to support decision-making in forest production planning, as well as their applicability in real-world scenarios.