Randomized algorithms for control and optimization
Rudolf Kulhavý · John Wiley & Sons, Inc. eBooks · 2000
A number of problems that industry and businesses are facing today are hard to deal with using analytical, calculus-based tools. Such problems include • estimation of parameters in nonlinear or otherwise complex models, • design of feedback control law in nonlinear control problems, • engineering design of complex and not completely understood manufacturing processes, • design of scheduling rules in ATM switches, • buffer allocation in production line, • estimation of rare event probability in high reliability systems, • routing policy in communication networks, • logistic system and policy design in global transportation network. The common characteristics of such problems are lack of structure (resulting from combination of combinatorial, discrete, and symbolic variables), inherent presence of uncertainties (requiring timeconsuming averaging), and huge search space (not easy to parameterize and prone to combinatorial explosion). Hard problems can be solved only approximately. The purpose of this paper is to show that randomized algorithms based on statistical simulation of variant models and designs yield a consistent and systematic framework for design and analysis of approximate solutions to such problems.