Models and Computational Intelligence Approaches for the Last Mile Delivery Problem in Food Supply Chain

Abstract

In cities, the "last mile" is not only a logistical issue, but also a significant urban planning challenge. The final mile in the supply chain involves highfrequency, low-volume, and short-haul distribution of products to end consumers. The last leg of the supply chain is the most crucial, but also the least efficient. Transportation planning is one of the major contributors to the severity of last-mile delivery (LMD) issues in cities. The scope of our thesis is on the transportation planning for perishable food supply chains. In recent years, the global food market has shown substantial growth. Therefore, posing new challenges, and initiating a drastic change in the last mile logistics. These challenges are mainly related to the high perishability of food product that require specific management approaches to maintain their quality. Additionally, we mention the growing demand from customers for deliveries to be made on time, at a lower cost, and in an environmentally friendly manner. Customers search other alternatives, if the service doesn’t meet their expectations. Last mile logistics is essential in developing brand loyalty in this competitive environment, as it allows transporters to deliver fresh and high quality products to consumers, faster and cost-effectively. The aim of this thesis is to propose models and solving approaches that can support decision-making process in food supply chain through addressing concerns about transportation costs, carbon emissions, product quality, service level, etc. Specifically, different decision problems that can be regarded as routing problems, are considered. We start by modelling a delivery problem with specific temporal customer’s requirements as a capacitated vehicle routing problem with soft time windows. The problem is formulated as a Mixed 0-1 non-Linear Program (MINLP). The objective is to minimize the total cost, consisting of transportation costs, food quality degradation costs, and time-windows violation costs. To model the temporal preference of the customers in a most realistic way, we provide another model that represent this preference information as a fuzzy number with respect to the satisfaction of service time. This problem is a Capacitated Vehicle Routing Problem with Fuzzy Time Windows (CVPFTW) and formulated as a fuzzy mixed-integer linear programming model. Both problems are addressed on real road networks where arcs are labelled with multiple attributes, and solved using CPLEX solver. The performance of the proposed models is assessed through computational analyses on several test instances. Some instances are derived from real-life applications, others are randomly generated. The results proved that our approach can help reduce the operational costs of delivery while improving customer service.To improve customer satisfaction, we propose a many-objective CustomerCentric Perishable Food Distribution Problem. The proposed model focuses on the cost, the quality of the product, and the service level improvement by considering not only time windows but also the customers’ target time and their priority. Owing the difficulty of solving such model, we propose a General Variable Neighbourhood Search (GVNS) metaheuristic. By solving a mono-objective sub-problem, such approach enables to generate a set of diverse solutions. These solutions are evaluated over some non-optimized criteria and then ranked using an a posteriori approach that requires minimal information about decision maker preferences. The computational results show: (a) GVNS achieved the same quality solutions as an exact solver (CPLEX) in the sub-problem; (b) GVNS can generate a wide number of candidate solutions, and (c) the use of the a posteriori approach makes easy to generate different decision maker profiles which in turn allows obtaining different rankings of the solutions. Solving a perishable food distribution problem in a real world setting is a very complex task. This is due to products characteristics, and the requirements of customers.To ensure a safe, quality product with a desired service level, a bunch of specifications should be included during the decision/optimization process. Many times, the computational models necessarily leave out of consideration several characteristics and features of the real world. Thus, trying to obtain the optimum solution can not be enough for a problemsolving point of view. To address this problem, we propose a modelling to generate Alternatives- metaheuristic based approach to generate a set of alternative solutions. The aim is to allow the decision maker to consider different perspectives, and non-modelled criteria.