ARTIFICIAL NEURAL NETWORK TECHNIQUE TO ESTIMATE FOREST EXTRACTION WORK CYCLE TIME IN A MOUNTAINOUS SITE

Forest harvesting is a complex activity, involving the movement of machines and wood volume being affected by several variables that interfere directly or indirectly in this forest operation. Linear models can be used to evaluate the impact of some of these variables on forest harvesting, although linear models have some limitations that prevent a better inference, for this reason, other alternatives such as artificial neural networks (ANN) can contribute to the understanding of the effect of variables on harvesting operations. The objective of this study was to compare the estimates of operational cycle time and the cycle elements in the extraction activity with a tractor winch in mountainous regions. Linear models were adjusted for each of the eight cycles evaluated (7 work steps and work cycle) in addition to seven neural network architectures for each cycle, totaling 56 trained architectures. The results show that the best neural networks trained for each work step presented superior adjustment statistics compared to linear models. In addition to superior results, the ANN presented normal residuals in most cases, a situation not achieved by linear models.