Sequence Based Memetic Algorithms for Static and Dynamic Travelling Salesman Problems

Hybridization of genetic algorithms (GAs) with local search techniques has received significant attention in recent years and is being widely used to solve real-world problems. These hybrid GAs, also called memetic algorithms (MAs), are able to incorporate other powerful techniques within the framework of GAs, working as a single unit and counterbalancing each other’s disadvantages. In this thesis, we propose a hybrid GA, called Sequence Based Memetic Algorithm (SBMA) with Inver Over (IO), for solving the travelling salesman problem (TSP). This is a 2-phase MA. The first phase (SBMA) consists of traditional binary operators, and the second phase is based on a unary operator. In SBMA, a tour is split into equal sub-tours. Further, the shortest tour is selected among the sub-tours and then optimized locally. The sub-tours are stored in the memory and then used to guide the evolutionary process via a kind of embedded local search. Additionally, we apply some techniques to adapt the key parameters based on whether the best individual within the population improves or not while also maintaining the diversity. After the first phase, the hybrid algorithm enters the second phase which is the Inver Over with elite population. Here, the IO is directed to get a clue from the elite population by adding and preserving good edges. We have also shown that the above approach can be extended to handle the dynamic TSP. The framework of our hybrid approach, along with the integration of the nearest neighbour list, applying 2-Opt local search on sub-tours and adaptive parameter control in the first phase, and the elite population with the rotating gene pool strategies in the second phase, works well for the DTSP. In order to test the performance of the proposed approach for the DTSP, experiments were carried out based on different DTSP test beds. From the study, it has been observed that the integrated heuristics or meta-heuristics are able to produce good-quality solutions for the DTSP. We also analysed the effect of the gene pool and immigrants generated with the nearest neighbour algorithm, which works well with all DTSP test instances, under different dynamics.