Performance of four meta-heuristics to solve a forestry production planning problem

Abstract

The use of artificial intelligence as a tool to aid in the planning of forest production has gained more and more space. Highlighting the metaheuristics, due to the ability to generate optimal solutions for a given optimization problem in a short time, without great computational effort. The present study aims to evaluate the performance of the metaheuristics Genetic Algorithm, Simulated Annealing, Variable Neighborhood Search and Clonal Selection Algorithm applied in a model of regulation of forest production. It was considered a planning horizon of 16 years, in which the model aims to maximize the Net Present Value (NPV), having as restrictions age of cut between 5 and 7 years and minimum and maximum logging demand of 140,000 and 160,000 m3, respectively. Different combinations of configurations were considered for each of the metaheuristics, 30-second processing time and 30 replicates for each configuration, all processing being performed in MeP - Metaheuristics for forest Planning software. The Simulated Annealing metaheuristic obtained the best results when compared to the others, reaching the minimum and maximum demand demanded in all tested configurations, in contrast, the Genetic Algorithm was the one with the worst performance. Thus, the capacity to use metaheuristics as a tool for forest planning is observed.