Solving Real-World Routing Problems using Evolutionary Algorithms and Multi-Agent-Systems.

This thesis investigates the solving of routing problems using Evolutionary Algorithms (EAs). Routing problems are known to be hard and may possess complex search spaces. Evolutionary algorithms are potentially powerful tools for finding solutions within complex search spaces. The problem investigated is the routing of deliveries to households within an urban environment; the most common instance of this problem is that of daily postal deliveries. A representation known as Street Based Routing (SBR) is presented. This is a problem representation that makes use of the real-world groupings of streets and houses. This representation is an indirect problem representation designed specifically for use with EAs. The SBR representation is incorporated within an EA and used to construct delivery routes around a variety of problem instances. The EA-based system is compared against a Travelling Salesman Problem (TSP) solver, and the results are presented. The EA-based system produces routes that are, on average, slightly longer than those produced by the TSP solver. Real-world problems may often involve the construction of a network of delivery routes that are subject to multiple hard and soft constraints. A Multi-Agent System (MAS)-based framework for building delivery networks is presented that makes use of the SBR-based EA presented earlier. Each agent within the system uses an EA to construct a single route. Agents may exchange work (via auctions or by directly negotiated exchanges), allowing the optimisation of their routes. It is demonstrated that this approach has much potential and is capable of constructing delivery networks meeting set constraints over a range of problem instances and constraint values.