Generátor instancí problému obchodního cestujícího

Abstract

This thesis deals with the design and implementation of a software tool for working with instances of the traveling salesman problem derived from real geographic inputs. The application allows locations to be specified on a map or imported in standard formats, builds a matrix of distances or travel times using the OpenRouteService platform and, when needed, can switch to a local Open Source Routing Machine instance or to a fallback spherical distance computation when routing data are unavailable. The optimization core supports a range of heuristics and metaheuristics including the LKH solver, and the result can be checked by visualization on a map.

The experimental part compares selected methods on three benchmark sets: a symmetric TSPLIB reference instance, an asymmetric TSPLIB reference instance, and a real-world instance composed of European capital cities. It illustrates differences in runtime, solution quality, and the stability of stochastic approaches. The findings are tied to a specific configuration and the tested instances, and do not imply a universal ranking of algorithms outside this implementation.

The aim is to provide a coherent tool for teaching and for studying algorithm behaviour on road-like data, while making clear the dependence on external services and on the deployment environment.