The Effects of Neighborhood Storage Size on Dynamic Programming Solutions
Abstract:Forward recursive dynamic programming (DP) was used to investigate the sensitivity of two- and three-state dynamic programming networks to neighborhood storage. Two-state DP networks were formed using all pairwise combinations of four state variables (basal area, cubic foot volume, number of trees, and diameter) and various neighborhood intervals for each state variable. As state neighborhoods decreased in size, all solutions were converged on a single rotation scheme and objective function value. Large state neighborhoods with nonuniform future growth and value potentials result in suboptimal solutions and incorrect sensitivity analyses. Comparison of various two-state networks indicate that models incorporating net cubic volume are less sensitive to increases in state neighborhood size. The addition of both a forward "look ahead" heuristic and the "harvest type" state variable to two-state networks improved the objective function value when neighborhoods were large and had nonuniform intra-neighborhood future growth and value potentials. However, these additional efforts resulted in no improvement in the objective function value when small state neighborhoods were used, further supporting the hypothesis that the DP algorithm did find the global optimum solution for the given problem formulation. For. Sci. 43(3):387-395.
Document Type: Journal Article
Affiliations: Associate Professor, University of Kentucky, Department of Forestry, Lexington, KY 40546-0073:, Fax: (606)323-1031
Publication date: August 1, 1997
More about this publication?
- Membership Information
- ingentaconnect is not responsible for the content or availability of external websites