%0 Research article
%T Population-based methods in the optimization of stand management
%A Pukkala, Timo
%D 2009
%J Silva Fennica
%V 43
%N 2
%R doi:10.14214/sf.211
%U https://silvafennica.fi/article/211
%X In Finland, the growth and yield models for tree stands are simulation  programs that consist of several sub-models. These models are often  non-smooth and non-differentiable. Direct search methods such as the  Hooke-Jeeves algorithm (HJ) are suitable tools for optimizing stand  management with this kind of complicated models. This study tested a new  class of direct search methods, namely population-based methods, in the  optimization of stand management. The tested methods were differential  evolution, particle swarm optimization, evolution strategy, and the  Nelder-Mead method. All these methods operate with a population of  solution vectors, which are recombined and mutated to obtain new  candidate solutions. The management schedule of 719 stands was optimized  with all population-based methods and with the HJ method. The  population-based methods were competitive with the HJ method, producing  0.57% to 1.74% higher mean objective function values than HJ. On the  average, differential evolution was the best method, followed by  particle swarm optimization, evolution strategy, and Nelder-Mead method.  However, differences between the methods were small, and each method  was the best in several stands. HJ was alone the best method in 7% of  stands, and a population based method in 3% (Nelder-Mead) to 29%  (differential evolution) of stands. All five methods found the same  solution in 18% of stands.