Since the performance of entire supply chain has a close relation with the structure and the way the supply chain components communicate, designing and managing the supply chain is one of the most vital issues for resource and service managers. Concentrating on the supply chain separately, leads to inefficiency and high cost of product and service delivery system. One of the main components of supply chain that has an important role in its designing and achieving competitive advantage for the organization is suppliers of raw materials. According to the importance of integrating decisions and selecting proper vendors in supply chain,we consider an integrated supply chain model that combines strategic decisions (selecting of proper vendors), with tactical decisions (production and inventory levels) and operational decisions (satisfying customer demand on time). Actually it is a multi-product vendor selection problem in a three level supply chain that contains several vendors, one manufacturer and one retailer which contains vendor selection decisions, assigning raw materials to vendors and inventory decisions of raw materials and final products simultaneously. We assume that raw materials arrive to the manufacturer immediately after purchasing and in each time period raw materials convert to finished products at the same time period and finally deliver to the retailer. A mathematical MINLP model is developed for this problem. We take into consideration capacity constraint for vendors and order quantity constraint for each raw material. In addition to selecting right vendors, assignment of raw materials to the vendors and how much to order to these selected vendors, the purpose of this model is integrating of inventory decisions in the three levels of supply chain. Two metahuristic approaches, Genetic Algorithm and Simulated Annealing, are proposed for solving this model. In order to evaluate the performance of these algorithms, the proposed model is also solved with GAMS software. Computational results show a significant decrease in the runtime of these two algorithms (about 89 percent for GA and 87 percent for SA) and achieving good solutions that are very close to the GAMS solutions (average 3.7 percent for GA and 4.2 percent for SA). According to the computational results we can claim that these two approaches are proper for solving the proposed model. Also it shows that with a little difference, GA has better performance than SA.
