Sensitivity Analysis and Real-Time Control of Discrete Event Systems

点击翻译按钮获取中文摘要

The two major objectives of this project are (a) to establish a general framework for estimating performance sensitivities of discrete event dynamic systems (DEDS) on-line, and (b) to integrate sensitivity information into optimization algorithms with real-time applicability. The proposed research will pursue some new directions in sample-path-based methodologies, using efficiently constructed "augmented systems." Augmented systems are intended to capture the stochastic behavior of both a nominal and perturbed DEDS, using only information directly observed. For a broad class of models and performance criteria, this framework is expected to provide sensitivity estimates with respect to parameters which may be real (e.g. job flows) or integer-valued (e.g. thresholds used in optimal dynamic policies). Algorithms will be developed for dealing with fundamental DEDS control problems, such as dynamic routing, scheduling, and adaptive resource sharing. Sensitivity analysis will be integrated into static and dynamic distributed algorithms for addressing performance optimization problems in the presence of real-time constraints (deadlines). The results of this project are expected to contribute to the growing effort towards general methodologies for DEDS analysis. They are also expected to provide workable algorithms for on-line optimization of complex systems with no available analytical models. The results from this work are very promising for dealing with control issues for such emerging technological environments as automated manufacturing, distributed processing systems, and computer networks. The PI is well qualified and the institution provides an excellent environment for the research. A three-year award is highly recommended.