Desenvolvimento de técnicas de inteligência artificial no manejo florestal

Abstract

The use of computational intelligence in the forestry sector is recurrent for several problems, mainly as an aid in decision making for forest managers. Congruent to this, the expansion of big data in the sector is notable, providing a large number of data. Thus, the application of methodologies that assertively facilitate the compilation and processing of this data becomes essential. Given this perspective, the thesis focuses on the use of the genetic algorithm to solve problems in the forestry sector, focusing on the selection and allocation of variables in a nonlinear tapering model (article 1), in addition to its application with a multiobjective character for selection of trees to be explored in the Amazon rainforest (article 2). In article 1, the objective of the work was to generate an efficient model for predicting the diameter along the shaft, in addition to investigating a biological approach to the selection of morphometric variables. Classical, morphometric, temporal and a constant variable were made available to the genetic algorithm in order to select and allocate these variables in the Kozak model (2004). In addition, the adjustments to the regression coefficients were performed by the algorithm. The model selected by the algorithm showed good diameter predictions along the shaft, with the selected variables having logical behaviors with the diametric increment. Article 2 was intended to use a multiobjective algorithm to assist in decision making in the selective cutting of a Mixed Ombrophilous Deciduous Forest, based on economic and environmental criteria. Which generated multiple solutions classified according to economic, ecological and balanced nature. The genetic algorithm was efficient in reconciling the different objectives, allowing low impact on the remaining stand. When analyzing the results presented in the thesis, it is noted that the genetic algorithm was effective in solving the proposed problems. In addition to its flexibility for different approaches, it is easy to structure for binary and continuous variables, and mono and multiobjective scenarios.