Modeling and evaluation of different optimization methods of forest assortment
Name: RODRIGO FREITAS SILVA
Publication date: 14/12/2018
Advisor:
Name |
Role |
|---|---|
| GILSON FERNANDES DA SILVA | Co-advisor * |
Examining board:
| Name |
Role |
|---|---|
| ADRIANO RIBEIRO DE MENDONÇA | Internal Examiner * |
| ANTONIO ALMEIDA DE BARROS JUNIOR | External Examiner * |
| GILSON FERNANDES DA SILVA | Co advisor * |
Summary: An increasingly competitive forest market linked to the demands of wood multiproducts favors the study of optimization methods that maximize the revenue of forest enterprises. Comparatively, little is known about the efficiency and effectiveness of the different solution methods applicable to the tree bucking optimization problem. The difficulty in finding the optimum assortments is due to the exponential growth of the number of cutting patterns to be analyzed in function of the number of products marketed and the stem dendrometer measurements. Therefore, some strategy is needed to optimize the search system. In this sense, the aim of this work was to model
mathematically the bucking optimization problem and evaluate through three case studies different proposed solution methods to solve this problem. The data corresponding to the case study 1 contains 408 Pinus taeda L. trees from Santa Catarina. The data from case study 2 correspond to 197 Eucalyptus sp. trees from the south of Bahia. The case study 3 is composed of 42.974 Eucalyptus saligna trees from Paraná. The optimization methods implemented at stem-level bucking-to-value analyzed by case studies 1 and 2 were: (1) Dynamic Programming (DP), (2) greedy algorithm, (3) exhaustive search, (4) Heuristic of Parts Construction (HPC) and the metaheuristics (5) Greedy Randomized Adaptive Search Procedure (GRASP) and (6) Iterated Local Search (ILS). The results of these algorithms were compared in the case study 1 with those results already known in the literature for Genetic Algorithm (GA)
and Simulated Annealing (SA). In case study 2 the market value of the logs was determined according to their quality class. Four quality classes were defined in function of the number of existing nodes. On the other hand, in the bucking-to-demand optimization system analyzed by case study 3, it was evaluated a multilevel optimization system implemented through the following hybrid solution methods: (1) DP + Integer Linear Programming (ILP) processed by CPLEX, (2) DP + Intensive Search Heuristic (ISH), (3) GRASP + ISH, (4) ILS + ISH and (5) HPC + ISH. Although DP has been the best solution method for solving the stem-level bucking optimization, HPC, GRASP and ILS achieved excellent results reaching, respectively, to 99,99%; 99,93% and 99,01% of the optimal solution in case study 1 and to 99,98%; 99,97%
and 99,84% of the optimal solution in the case study 2. In bucking-to-demand system (case study 3), the DP + ISH was able to reach 99,77% of the optimal solution in less than half the time spent by DP + ILP to reach the exact solution of the problem.
Keywords: Operational research, heuristics, multiproducts.
