Abstract— This paper focuses on the second stage of a three- stage, integrated methodology for modeling and optimisation of distribution networks based on Hybrid Genetic Algorithms. The methodology permits the use of any combination of transportation and warehousing costs for a deterministic demand. This paper analyses and compares the variation of overall costs when the number of facilities varies and indicates how to minimize them. The distribution network directly and critically affects costs, efficiency and service level - the essential performance operation indicators for supply chains. The paper concentrates on Capacitated Location Allocation of distribution centers, a large scale, highly constrained, NP- hard, combinatorial problem. The Hybrid Genetic Algorithm used has a classical structure, but incorporates a special encoding of solutions as chromosomes and the integration of a Linear Programming/Mixed Integer Programming module embedded in the generation, crossover and pseudo-mutation operators. A complex and extensive case study – 25 production facilities, 5 to 10 distribution centres and 25 retailers (up to 520 variables intricately connected with a significant number of constraints) – is described, demonstrating the robustness of the Hybrid Genetic Algorithm and the optimization approach. Index Terms— Distribution network, Capacitated Location Allocation Problem, Optimisation, Hybrid Genetic Algorithms. I. INTRODUCTION he distribution network directly and critically affects the structure, complexity, costs and overall efficiency associated with operating a Supply Chain (SC), as well as the service level, the two critical factors in any SC [1]. This directly influences the aptitude of participants in a SC (whether is under control of a single or a conglomerate of companies) to enter or stay competitive in a market. An aspect of the utmost importance is the escalating complexity of distribution networks. As SC become increasingly large and complex, a general trend today, due mainly to globalization [2], designing the distribution network becomes vital [3], [4]. Due to sheer size, capacity of classical tools to solve and optimise LA problems in real distribution networks was exceeded. For this reason, Manuscript received December 2010; revised February, 2011. Romeo MARIAN, PhD is Senior Lecturer and Program Director at School of Advanced Manufacturing & Mechanical Engineering, University of South Australia, Australia (e-mail: romeo.marian@unisa.edu.au). Son Duy DAO is postgraduate student at School of Advanced Manufacturing & Mechanical Engineering, University of South Australia, Australia (corresponding author, e-mail: daosd001@mymail.unisa.edu.au). Professor Lee Luong is Head of School of School of Advanced Manufacturing & Mechanical Engineering (email: lee.luong@unisa.edu.au) development of new tools and implementations were and are necessary. The paper presents a framework developed for the optimisation of distribution networks in SC, especially as the Location-Allocation (LA) of distribution centres (DC) or warehouses are concerned. The size of DC/warehouses is limited – thus, the problem becomes a Capacitated Location Allocation (CLA) problem. The CLA problem is a large scale, highly constrained, NP-hard, combinatorial problem [5]. Genetic Algorithms (GA) have been chosen in this research for the design and optimisation of the distribution network for their remarkable capacity to successfully work with problems with huge solution spaces [6]. Various GA have been developed, as shown briefly in the next section, to solve a plethora of LA - like problems. Some of them were successful, at least to a certain degree, to tackle - in general - restricted variants of the problem. This becomes obvious when the run-time of a genetic algorithm starts clocking hours of CPU and indicates that particular methodology has attained its limits, mainly due to the general foe of combinatorial optimisation, the combinatorial explosion. The approach presented in this paper to tackle the CLA problem differs from others through the use of an integrated methodology, flexible enough to accommodate most realistic assumptions and, at the same time, computational- resource conscientious, to avoid combinatorial explosion. The methodology presented here is a development of algorithms of similar complexity presented in [7], [8] and an extension of the work in [3], [4]. A complex case study, of considerable size and a complex cost structure, achieved excellent run times. It demonstrates the robustness of the Hybrid Genetic Algorithm (HGA) and its capacity to tackle problems of considerable size and the expandability of the HGA to even larger and more complex problems. II. LITERATURE REVIEW ON OPTIMISATION OF LOCATION-ALLOCATION PROBLEMS The LA problems received considerable attention in the last 50 years due to their immediate practical applicability. A wide variety of methodologies and techniques were employed, alone or in diverse combinations in attempting to find good solutions. LA problem were treated in depth in recent years [9], [10]. Variations of the problem appear in numerous papers. Different definitions of variants of LA problems make in fact difficult to compare results between different problems. The literature review presented here shows mainly the Modeling and Optimisation of Distribution Networks Using Hybrid Genetic Algorithms: A Comparative Study Romeo M. Marian, Lee H.S. Luong and Son Duy Dao T