随机规划:
http://stoprog.org/index.html?SPTutorial/SPTutorial.html
This example raises several questions. First, how should we approximate the random variable by one with a finitely-supported probability distribution? Second, how should we solve the resulting (approximate) optimization problem? Third, how can we gauge the quality of the approximate solution? In the next section we will discuss these issues in the context of two-stage linear stochastic programming problems. One natural generalization of the two-stage model, which we discuss in section 3, extends it to many stages. Here each stage consists of a decision followed by a set of observations of the uncertain parameters which are gradually revealed over time. In this context stochastic programming is closely related to decision analysis, optimization of discrete event simulations, stochastic control theory, Markov decision processes, and dynamic programming. Another class of stochastic programming models seeks to safeguard the solution obtained against very bad outcomes. Some classical approaches to deal with this are mean-variance optimization, and so-called chance constraints. We briefly discuss these in section 4, and explore the relationship with their modern counterparts, coherent risk measures and robust optimization.