This paper proposes an agent-based metaheuristic to solve large-scale multi-period supply chain network design problems. The generic design model formulated covers the entire supply chain, from vendor selection, to production–distribution sites configuration, transportation options and marketing policy choices. The model is based on the mapping of a conceptual supply chain activity graph on potential network locations. To solve this complex design problem, we propose Collaborative Agent Team (CAT), an efficient hybrid metaheuristic based on the concept of asynchronous agent teams (A-Teams). Computational results are presented and discussed for large-scale supply chain networks, and the results obtained with CAT are compared to those obtained with the latest version of CPLEX.
In recent years, the emphasis on trade globalization as well as the emergence of new economic powers such as the Brazil, Russia, India, and China (BRICs) brought forth new competitive challenges as well as new opportunities for growth and cost reductions. The ensuing mergers, acquisitions as well as supply chain reconfigurations involve a large number of complex inter-related supply chain network (SCN) design decisions that heavily impact company's competitive position, debt and profitability. Moreover, the large investments associated with these decisions require the consideration of a planning horizon covering several years. In such a context, companies seek to improve their profitability by generating economies of scale as well as making efficient use of capital while improving customer service (Cooke, 2007). Given the complexity and interdependence of supply chain network design decisions, it has been shown that the use of operations research techniques and tools such as mixed-integer programming models can result in significant returns (Geoffrion and Powers, 1995 and Shapiro, 2008). Unfortunately, the problems to be modeled are so large and complex that even the best-of-breed commercial solvers are seldom able to solve real instances to optimality in a reasonable amount of time. Thus, the need for an efficient and flexible heuristic solution method arises.
A typical SCN design problem sets the configuration of the network and the missions of its locations. Some facilities may be opened, others closed, while others can be transformed using different capacity options. Each selected facility is assigned one or several production, assembly and/or distribution activities depending on the capacity options available at each location. The mission of each facility must also be specified in terms of product mix and facilities/customers to supply. Key raw-material suppliers must be selected. For each product-market, a marketing policy setting service and inventory levels, as well as maximum and minimum sales levels, must also be selected. The objective is typically to maximize net profits over a given planning horizon. Typical costs include fixed location/configuration costs, fixed vendor and market policy selection costs, as well as some variable production, handling, storage, inventory and transportation costs (Amrani et al. 2011).
The objective of this paper is, first, to propose a generic formulation of the multi-period SCN design problem based on the mapping of a conceptual supply chain activity graph on potential network locations, and, second, to propose an efficient hybrid metaheuristic based on a collaborative agent team (CAT) to solve large instances of this model. The rest of the paper is organized as follows. In Section 2, a general review of the relevant literature is provided. Section 3 defines the activity-based concepts required to model SCNs. Section 4 formulates the mathematical programming model to be solved. Section 5 outlines the solution approach developed to tackle the problem. Computational results are presented and discussed in Section 6, and Section 7 concludes the paper.
This paper proposed a novel modeling approach for activity-based multi-period supply chain network design problems. It effectively integrates design and modeling concepts found in previous papers into a generic model that can be efficiently used to reengineer real-world supply chain networks. An agent-based metaheuristic (CAT), grounded in the A-Teams paradigm, was also proposed to solve this model effectively. Comparisons with CPLEX indicate that our algorithm performs better on the vast majority of the instances solved and for all problem structures. Furthermore, by using a metaheuristic such as CAT, one is not forced to use linear constraints and objectives (or approximate nonlinearities by piecewise linear equations). This opens up new modeling opportunities. Furthermore, the CAT metaheuristic can easily be extended and improved by adding new agents as needed.
There are two main avenues to extend this work. From a CAT implementation perspective, much could be done to increase the efficiency of agents and reduce the time spent on nonproductive tasks such as writing and reading solutions. From the SCN modeling point of view, the model presented could be extended to incorporate financial constraints, international factors and reverse logistics structures. Finally, in order to account for the uncertainty inherent in these multi-period problems, a scenario-based stochastic programming version of the model could and should be elaborated.