An improved simulated annealing algorithm for solving spatially explicit forest management problems.

摘要

Spatially-explicit forest management problems that explicitly convey location information offer many new opportunities. The greatest opportunity is the ability to incorporate new constraints, such as adjacency constraints, that are difficult to specify without having spatially-explicit decision variables. These constraints either prevent the simultaneous harvest of adjacent management units (unit restriction model, or URM) or prevent the simultaneous harvest of contiguous forest areas that exceed a certain prescribed limit (area restriction model, or ARM). Such constraints are either legally required or voluntarily implemented in many situations in the US. Solving spatially-explicit forest management problems in a reasonable amount of time is an important challenge.; Heuristic techniques have been used to reduce solution times for spatially-explicit forest management problems. In particular, simulated annealing (SA) is a heuristic algorithm that has been widely accepted by the scientific community as it has been shown to be at least theoretically capable of finding an optimal solution. This study demonstrates methods that can improve the performance of SA by: (1) optimizing the SA parameters, and (2) improving the algorithm so that variables are selected to be added to the solution with probability related to their potential to generate improved solutions (which is measured in terms of a simple index which we call desirability) rather than with equal probability. This new algorithm is referred to as SAPRD.; Chapter 2 presents a method for optimizing the SA parameters and the results of an experiment to test various hypotheses about the estimated optimal parameters. This experiment showed that there is no significant difference between the performance of SAPRD with a parameter set estimated for a particular class of problems and with a parameter set estimated for an individual problem within that class. Similarly, there was no significant difference in the performance of SAPRD with parameter sets optimized for different solution times. However, the performance of SAPRD improved significantly when longer solution times are actually used for solving problems.; The new algorithm, SAPRD, is discussed in Chapter 3 along with two new strategies for fast selection of units with probability related to their desirability. SAPRD uses a simple, easy-to-implement method for assigning different probabilities to the units that is based on an index that is calculated from the objective function coefficients and the technical coefficients of the constraints. (Abstract shortened by UMI.)