Publication: A parallel implementation of the Dynamically Dimensioned Search (DDS) algorithm
All || By Area || By Year| Title | A parallel implementation of the Dynamically Dimensioned Search (DDS) algorithm | Authors/Editors* | Tolson, B. A., Sharma, V. and Swayne, D. |
|---|---|
| Where published* | International Symposium on Environmental Software Systems (ISESS) 2007 |
| How published* | Proceedings |
| Year* | 2007 |
| Volume | |
| Number | |
| Pages | |
| Publisher | International Symposium on Environmental Software Systems |
| Keywords | |
| Link | |
| Abstract |
This paper introduces two parallel implementations of a recently developed global optimization algorithm developed by Tolson and Shoemaker (2007), Dynamically Dimensioned Search (DDS). DDS is a simple, robust tool for the automatic calibration of computationally expensive and highly parameterized environmental simulation models. DDS is unique because the simple and parsimonious (single parameter) algorithm structure enables it to work well when subjected to computational limitations. In other words, DDS was designed from the perspective that the total number of model evaluations utilized for model calibration is always limited and case study dependent. The extension to parallel computing means that wall clock time is also limited. Parsimony of structure is possible because DDS was designed to find practical, acceptable quality solutions for the calibration problem rather than a globally optimal one. Results are presented comparing the serial DDS algorithm to an âEmbarrassingly Parallelâ DDS algorithm (EP-DDS) as well as a parallel DDS (P-DDS) algorithm with increased processor communication. The SAC-SMA hydrologic model is used, with 13 calibration parameters to assess the effectiveness of the parallel DDS algorithm as the numbers of processors are increased. Results also indicate that either of the parallel versions of the DDS algorithm is a promising extension to the serial version of DDS. |
Back to page 62 of list