Μεθοδολογίες βελτιστοποίησης για μεγάλης κλίμακας προβλήματα δρομολόγησης στόλου οχημάτων

Abstract

In the present PhD thesis, a collection of vehicle routing problems is examined and solved. These problems are aimed at producing the highest quality routes for performing various transportation operations. They frequently arise in practical applications of Logistics, and therefore, are of great importance both from the theoretical and commercial perspectives. For this reason, vehicle routing problems have attracted the attention of researchers from various fields of science, namely Operations Research, Mathematics, Computing Science, and Management Science. Vehicle routing problem variants belong to the category of NP-hard combinatorial optimization problems, thus they are very complex to be solved. Specifically, exact mathematical algorithms cannot optimally solve practical, large-scale routing problems within reasonable computational times. For this reason, researchers focused their interest in developing heuristic algorithms which are problem-specific methodologies aimed at determining good solutions, not necessarily the optimal ones, within limited computational times. Later, researchers introduced a new generation of approximate algorithms called metaheuristics which are able to produce higher quality solutions compared to the ones obtained by classical heuristics. Metaheuristics are intelligent strategies which incorporate low-level heuristic methods into algorithmic frameworks which are aimed at effectively exploring the solution search space. The basic attribute of metaheuristic strategies is that their behaviour does not depend on the special characteristics of the examined problem. Thus, they provide great flexibility of being applied for dealing with problems of diversified characteristics. The present PhD thesis proposes a collection of innovative metaheuristic methodologies for effectively tackling various problems of the vehicle routing literature. In methodological terms, a strategy for reducing the computational complexity of local search-based methods is introduced. Various metaheuristic schemes based on the aforementioned strategy are presented. The proposed complexity reduction strategy can be employed for dealing with very large-scale applications of vehicle routing problems. In addition, it can be used for efficiently applying rich and complex local search operators. It can also drastically accelerate the application of local search methods which are the most common algorithmic feature of commercial vehicle routing software packages. Moreover, the present thesis introduces several hybrid approaches which combine the powers of more than one metaheuristic method. In specific, two hybrid schemes are proposed which jointly employ the rationale of Tabu Search and Guided Local Search. Furthermore, an innovative metaheuristic mechanism which coordinates the performance of local search methods is introduced and analyzed. This mechanism is called Promises and exhibits a considerably robust behaviour, as it does not include any algorithmic parameters. Finally, various original methodological schemes are proposed, for tackling vehicle routing problems integrated with additional loading constraints. These schemes involve hybrid metaheuristic methods for dealing with the routing requirements operating in parallel with packing heuristic methods aimed at determining feasible packing arrangements for the transported products. The abovementioned algorithms are applied to deal with several vehicle routing variants. Specifically the following models are solved: (a) the classical Capacitated Vehicle Routing Problem, (b) the Open Vehicle Routing Problem, (c) the Vehicle Routing Problem with Backhauls, (d) the Vehicle Routing Problem with Simultaneous Pick-ups and Deliveries, (e) the Undirected Capacitated Arc Routing Problem with Profits, (f) the Capacitated Vehicle Routing Problem with Two-Dimensional Loading Constraints, and (g) the Capacitated Vehicle Routing Problem with Three-Dimensional Loading Constraints. The proposed algorithms were evaluated by being applied on benchmark instances for all the aforementioned models. They exhibited fine performance by improving most of the best known solutions of the literature. Finally, in the present PhD thesis, a new vehicle routing problem integrated with additional loading constraints is introduced, formulated and solved. This new problem models applications of transporting palletized products, and generalizes every integrated routing-packing model of the literature. To solve this introduced vehicle routing problem, an implementation of the Promises mechanism is proposed which operates in parallel with a bundle of 144 three-dimensional packing heuristics aimed at determining feasible loading structures for the transported products onto the vehicle pallets.