MIT教材，系统优化与 及调度ppt 使用美国麻省理工的教材，系统优化与 及调度，权威ppt Modeling Optimization of Complex Networked Systems Applications to Operations Scheduling Supply Network Management.ppt

9/14/2001 Fall 2001, Copy Right P. B. Luh * Reading Assignment: Bertsekas Sections 12 - 1.4, and Appendix C Problem Set 1: 1.1.3, 1.1.4, 1.2.1, 1.2.9, 1.3.1, 1.3.5, 1.4.3 Web: /lectures_cfins.html or 2/CFINSCourseOutline01FLuhR4.htm Last Time: Introduction and Unconstrained Optimization Motivation and Course Overview Problem Classification Optimality Conditions for Unconstrained Optimization Gradient Methods ? Framework Today: One-dimensional Search Methods Gradient Methods ? Convergence and Rate of Convergence Newtons Method and Variations Next Time: Bertsekas Sections 1.6, 2.1 - 2.3 Modeling Optimization of Complex Networked Systems: Applications to Operations Scheduling Supply Network Management Iterative optimization: Minimize f(x), with x ? X ? Rn Start with an initial point x0, evaluate f(x0), ?f(x0) (and possibly ?2f(x0)) Based on the information, select a search direction “d” emanating from x0 Find the next point x1 along the direction d, and evaluate f(x1), ?f(x1) (and possibly ?2f(x1)), then repeat Key Questions: Which direction to go? How far? Is the algorithm guaranteed to reach x*? How fast to reach x*? x2 x3 x1 xk+1 = xk + ?k dk as long as ?f(x) d 0 f(xk+1) = f(xk) + ?k?f(xk)d + o(?k) f(xk) for small ?k A specific form: dk = - Dk ?f(xk) with Dk 0 ? Gradient methods How far to go along the direction d, or how to select the step size? What are the advantages and disadvantages of each method? 1.5 One-dimensional Search Methods There are several possibilities Minimization Approximate Stepsizing ? the Armijo Rule Diminishing Stepsize (Read yourself) A key component in multi-dimensional optimization Need to trade off between solution accuracy vs. computational efforts Minimization Minimizing f(x) along the dk direction ? “Line Search.” ?k = arg min ?? 0 f(xk + ? dk), Suppose f ? C1 (continuously differentiable), g(?) ? f(xk + ? dk), and g(?) = ?f(xk + ? dk) dk. Assume that g(0) = ?f(xk)dk 0, i.e., dk is a desce