Doktorska disertacija
Stigmergija kot pristop k metahevristični optimizaciji
Ogledi: 9 | Prenosi: 8
Povzetek
Developing metaheuristics to solve optimization problems is a rapidly growing field
of research. This is due to the importance of optimization problems in the scientific
as well as the industrial world. The methods developed in this dissertation are based
on stigmergy: a method of communication in emergent systems, where the individual
parts of the system communicate with one another by modifying their local environ-
ment. Stigmergy is the basis for all ant-colony optimization methods.
The introduction describes the components and concepts of Ant-Colony Optimiza-
tion, and shows how ant-colony optimization methods have been used to solve com-
binatorial optimization problems like ordering, assignment, subset and grouping. In
contrast, there have only been few attempts to solve numerical optimization problems
in this way.
Almost all algorithms for combinatorial optimization are based on a single ant
colony, while here we show how multiple ant colonies can be used to solve a grouping
problem called the mesh-partitioning problem.
There is no straightforward way to implement ant-colony optimization to solve a
numerical optimization problem. The main part of this dissertation is the develop-
ment of two algorithms for solving numerical optimization problems. With the first
one, called the Multilevel Ant-Stigmergy Algorithm, the problem is put into a discrete
form, which is also a typical strategy for other ant-colony optimization algorithms.
However in the second algorithm, called the Differential Ant-Stigmergy Algorithm,
this is no longer required, making the Differential Ant-Stigmergy Algorithm much
more versatile.
Both algorithms are evaluated on benchmark functions and compared to several other metaheuristic algorithms. They were also evaluated and compared on real-world
problems from the fields of electrical engineering and metallurgical production.
The dissertation ends with the conclusion that for most numerical optimization
problems addressed the Differential Ant-Stigmergy Algorithm is more suitable than
the Multilevel Ant-Stigmergy Algorithm, and some suggestions for future work.