Plenary Lecture 5: Integrated and real-time optimization: models, algorithms and applications

摘要

A major challenge in supply chain management is the development of computational models and methods for integrated and real-time optimization at the operational level. Companies are facing constant pressure to reduce their costs and improve customer service. As a consequence, they are forced to keep low inventory levels across the supply chain, to increase their responsiveness to customers, and globally to improve their operational performance. To achieve these goals, it is fundamental that planning and scheduling is made in an integrated and coordinated way across the different functions of the companies from procurement to delivery planning. Real-time optimization is used to address the inherent variability that characterizes any real system. It involves recomputing the plans and schedules with a high frequency to account for unpredicted events and new data that becomes available during execution. The goal is to reduce the planning cycles to incorporate real-time data, and respond effectively to new requests without compromising customer service. The potential benefit of integrated and real-time optimization is widely recognized. Despite this clear practical relevance, research on integrated and real-time optimization at the operational level is very recent, and it focused essentially on heuristic methods. Mixed Integer Programming (MIP) can be very valuable tool in this context. MIP has gone through major developments in the last decades. Combined with the constant progress in hardware, MIP has been applied successfully to the resolution of complex and large size problems in many sectors, such as the telecommunications and the commercial airline industry. Recently, different authors have reported successful applications of MlP-based methods to specific integrated and real-time optimization problems. In this talk, MlP-based approaches for these optimization problems will be discussed. Additionally, we will explore a new modeling technique that leads to models with a pseudo-polynomial size that can be managed dynamically so as to handle efficiently the spatial and temporal integration that characterize these problems.