A genetic algorithm approach to tree bucking optimization

Tree bucking on cut-to-length harvesters is controlled by two types of matrices. The price matrix provides the bucking computer with information on how to prioritize various log grades and diameter–length combinations within the same grade, while the demand matrix specifies the desired proportion for each combination. The traditional approach has been to apply the same matrix set to all stands to be harvested within the same planning period, although the stand structure and the characteristics of trees may vary markedly from stand to stand. The purpose of this research was to test the hypothesis that controlling bucking matrices prior to harvesting would improve bucking results at the forest level. The search for stand-specific prices matrices was based on a genetic algorithm (GA) that, given a desired overall log distribution and the stem profiles of all the trees in each stand, optimizes, in a parallel manner, the price matrix of a given log grade for each stand involved in the process. The mutation rate and the degree of elitism had the greatest effect on the performance of the developed GA system. In 10 test runs with the same parameter set, the fitness value of the poorest solution (price matrix string) was 98.2%) of that of the best solution. The simulations with the bucking simulator, however, indicate that precontrol of price matrices does not improve the fit between the overall demand matrix and the global output matrix even if the log prices are adjusted according to stem data without estimation errors.