The uses of the slime mold lifecycle as a model for numerical optimization.

详细信息   

• 作者：Monismith ; David R. ; Jr.
• 学历：Doctor
• 年：2010
• 导师：Mayfield, Blayne E.,eadvisorChandler, John P.ecommittee memberHeisterkamp, Douglas R.ecommittee memberHagan, Martin T.ecommittee member
• 毕业院校：Oklahoma State University
• Department：Computer Science
• ISBN：9781124211008
• CBH：3422282
• Country：USA
• 语种：English
• FileSize：7830996
• Pages：188

文摘

Scope and Method of Study. This work provides a discussion of the lifecycle of the cellular slime mold, Dictyostelium discoideum Dd), as it may be used for numerical optimization with emphasis on its use as an Evolutionary Algorithm. The study begins with a review of a number of existing numerical optimization algorithms that make use of direct search methodology i.e. they do not require the computation of a derivative to perform optimization) such as Pattern Search, Downhill Simplex, and Razor Search. These algorithms are of interest because they were precursors to Evolutionary Optimization, and their search strategies, in some cases, are similar to amoeboid movement. Next, a review of some existing Evolutionary Algorithms is provided. This includes a review of Differential Evolution, Particle Swarm Optimization, and a Real-Coded Genetic Algorithm. The second part of the review is of Dd lifecycle, biological computation, and simulations thereof. With simulations in hand, several data structures are introduced to handle the transition from simulation to optimization. Then, the Slime Mold Optimization Algorithm is introduced. It follows the lifecycle of Dd, using vegetative, aggregative, mound, slug, and dispersive states to perform optimization. Thereafter, several variants of the Slime Mold Optimization Algorithm are created. Findings and Conclusions. The Slime Mold Optimization Algorithm and its variants were tested on a comprehensive function suite consisting of objective functions of varying difficulty, dimensionality, and modality. Results were compared by varying parameters of the algorithm including number of amoebae and maximum numbers of objective function values. Results were also compared to those of existing Evolutionary Algorithms. These results show promise and in some cases are better than existing Evolutionary Algorithms, though work is needed to make the algorithm better suited to extremely large search spaces and problems with high dimensionality. Variants of the algorithm were also tested showing improvement over the original version of the Slime Mold Optimization Algorithm.