Simulation-based optimization: Parametric optimization techniques and reinforcement learning (Q1869929)

The aim of the book is to suggest simulation-optimization techniques for solving stochastic-optimization problems that can be solved with computer simulation. Such problems are referred to as simulation-optimization problems. Both static parametric optimization and dynamic control optimization are considered in this context. The focus of this book is on complex, large-scale, stochastic systems, where a simple application of classical methods does not work and it is necessary to use a simulation-based evaluation of the objective function. It is necessary therefore to use nonlinear programming methods that rely only on functional evaluations. Further, the author presents some elementary mathematical facts, which prove that the optimization algorithms presented in the book produce optimal or in some cases near-optimal solutions. The book is divided into 16 chapters. The first three chapters have an introductory character. Chapters 4 and 5 cover basic concepts related to discrete-event simulation and present an overview of optimization with simulation. The response surface methodology used in conjunction with simulation optimization, the main techniques for parametric optimization with simulation, and the classical theory of stochastic dynamic programming are presented in Chapters 6-8. Reinforcement learning and solving Markov and semi-Markov decision problems are contained in Chapters 9 and 10. Some convergence results and the necessary mathematical analysis are contained in Chapters 11-13. Chapter 14 presents an overview of some case studies, Chapter 15 contains some computer programs of the algorithms from the previous parts of the book and Chapter 16 contains a concluding survey of the results achieved and challenges for further research in simulation optimization.