Optimizing heuristic search in forest planning

Heuristic search methods are being used more and more in forest planning since the current formulations of exact methods such as linear programming are not suitable to all today's planning problems. A practical problem with most heuristics is that their performance greatly depends on the parameters that guide their search process. The effect of parameters is hard to know without extensive tests, but these tests cannot be conducted in forest planning practice, because of lacking time and experience. This study presented a method that uses Hooke and Jeeves direct search to optimize the parameters of a heuristic, taking into account the allowed computing time. The method was used to optimize three local-improvement heuristics in a non-spatial and a spatial forest planning problem, and with a short and long computing time. The heuristics were simulated annealing, threshold accepting, and tabu search, all of which are used in forestry. The results were logical and showed that while the optimal values of some parameters were rather constant the others were sensitive to problem type, allowed computing time, or problem size. The objective function value of the forest planning problem was not sensitive to small changes in the parameters of the heuristics. However, because computing time was very sensitive to many parameters, there was not much freedom to set the parameters if both the quality of the solution and speed of the algorithm had to be maintained.