Stochastic Programming Project Konrad Borys. Model for Optical Fiber Manufacturing

Similar documents
CAPITAL ASSET PRICING MODEL (CAPM)

8 Laplace s Method and Local Limit Theorems

4.1. Probability Density Functions

Lecture 21: Order statistics

Quadratic Forms. Quadratic Forms

2D1431 Machine Learning Lab 3: Reinforcement Learning

LECTURE NOTE #12 PROF. ALAN YUILLE

3.4 Numerical integration

Administrivia CSE 190: Reinforcement Learning: An Introduction

1B40 Practical Skills

Can the Phase I problem be unfeasible or unbounded? -No

Section 11.5 Estimation of difference of two proportions

University of Texas MD Anderson Cancer Center Department of Biostatistics. Inequality Calculator, Version 3.0 November 25, 2013 User s Guide

Math 426: Probability Final Exam Practice

Best Approximation in the 2-norm

Math 270A: Numerical Linear Algebra

Chapter 9: Inferences based on Two samples: Confidence intervals and tests of hypotheses

Joint distribution. Joint distribution. Marginal distributions. Joint distribution

Continuous Random Variables

19 Optimal behavior: Game theory

Solution for Assignment 1 : Intro to Probability and Statistics, PAC learning

Math 135, Spring 2012: HW 7

Bellman Optimality Equation for V*

Where did dynamic programming come from?

Review of Probability Distributions. CS1538: Introduction to Simulations

Chapter 3 Solving Nonlinear Equations

Problem Set 9. Figure 1: Diagram. This picture is a rough sketch of the 4 parabolas that give us the area that we need to find. The equations are:

Integral equations, eigenvalue, function interpolation

We will see what is meant by standard form very shortly

Engineering Analysis ENG 3420 Fall Dan C. Marinescu Office: HEC 439 B Office hours: Tu-Th 11:00-12:00

Chapter 5 : Continuous Random Variables

Probability Distributions for Gradient Directions in Uncertain 3D Scalar Fields

Reinforcement learning II

Discrete Least-squares Approximations

Acceptance Sampling by Attributes

Numerical Analysis. 10th ed. R L Burden, J D Faires, and A M Burden

Pre-Session Review. Part 1: Basic Algebra; Linear Functions and Graphs

ODE: Existence and Uniqueness of a Solution

CS 109 Lecture 11 April 20th, 2016

38.2. The Uniform Distribution. Introduction. Prerequisites. Learning Outcomes

Line Integrals. Partitioning the Curve. Estimating the Mass

Section 4.8. D v(t j 1 ) t. (4.8.1) j=1

Trapezoidal Rule, n = 1, x 0 = a, x 1 = b, h = b a. f (x)dx = h 2 (f (x 0) + f (x 1 )) h3

Numerical integration

Lecture 3 Gaussian Probability Distribution

Lecture 12: Numerical Quadrature

Tests for the Ratio of Two Poisson Rates

Monte Carlo method in solving numerical integration and differential equation

Orthogonal Polynomials and Least-Squares Approximations to Functions

CBE 291b - Computation And Optimization For Engineers

Properties of Integrals, Indefinite Integrals. Goals: Definition of the Definite Integral Integral Calculations using Antiderivatives

Space Curves. Recall the parametric equations of a curve in xy-plane and compare them with parametric equations of a curve in space.

{ } = E! & $ " k r t +k +1

Section 3.2 Maximum Principle and Uniqueness

Numerical Analysis: Trapezoidal and Simpson s Rule

Chapter 4: Dynamic Programming

Stochastic Optimization: The Present and Future of OR

Energy Bands Energy Bands and Band Gap. Phys463.nb Phenomenon

ODE: Existence and Uniqueness of a Solution

Math 1B, lecture 4: Error bounds for numerical methods

Recitation 3: More Applications of the Derivative

Section 6.1 Definite Integral

INTRODUCTION TO INTEGRATION

Continuous Random Variables

Point Processing of Images. Point Processing of Images. EECE/CS 253 Image Processing. Point Processing

Suppose we want to find the area under the parabola and above the x axis, between the lines x = 2 and x = -2.

Intensity transformations

CS667 Lecture 6: Monte Carlo Integration 02/10/05

Numerical Integration

Numerical Integration. 1 Introduction. 2 Midpoint Rule, Trapezoid Rule, Simpson Rule. AMSC/CMSC 460/466 T. von Petersdorff 1

Math 116 Calculus II

Module 11.4: nag quad util Numerical Integration Utilities. Contents

Problem Set 3 Solutions

III. Lecture on Numerical Integration. File faclib/dattab/lecture-notes/numerical-inter03.tex /by EC, 3/14/2008 at 15:11, version 9

Section 6: Area, Volume, and Average Value

Bayesian Networks: Approximate Inference

x = a To determine the volume of the solid, we use a definite integral to sum the volumes of the slices as we let!x " 0 :

Online Supplements to Performance-Based Contracts for Outpatient Medical Services

CS103B Handout 18 Winter 2007 February 28, 2007 Finite Automata

Integrals along Curves.

Problem. Statement. variable Y. Method: Step 1: Step 2: y d dy. Find F ( Step 3: Find f = Y. Solution: Assume

Population bottleneck : dramatic reduction of population size followed by rapid expansion,

Tutorial 4. b a. h(f) = a b a ln 1. b a dx = ln(b a) nats = log(b a) bits. = ln λ + 1 nats. = log e λ bits. = ln 1 2 ln λ + 1. nats. = ln 2e. bits.

Chapter 3 Polynomials

Best Approximation. Chapter The General Case

Chapter 2 Organizing and Summarizing Data. Chapter 3 Numerically Summarizing Data. Chapter 4 Describing the Relation between Two Variables

Session Trimester 2. Module Code: MATH08001 MATHEMATICS FOR DESIGN

Reinforcement Learning

1 Linear Least Squares

Expectation and Variance

Travelling Profile Solutions For Nonlinear Degenerate Parabolic Equation And Contour Enhancement In Image Processing

Review of Gaussian Quadrature method

Generalized Fano and non-fano networks

Construction of Gauss Quadrature Rules

Lecture 3 ( ) (translated and slightly adapted from lecture notes by Martin Klazar)

Heat flux and total heat

Chapter 6 Continuous Random Variables and Distributions

4.5 JACOBI ITERATION FOR FINDING EIGENVALUES OF A REAL SYMMETRIC MATRIX. be a real symmetric matrix. ; (where we choose θ π for.

A Matrix Algebra Primer

Chapter 2 Fundamental Concepts

Transcription:

Stochstic Progrmming Project Konrd Borys Model for Opticl Fiber Mnufcturing. Introduction Opticl fibers re mde of solid rods of glss clled preforms. he s of the preforms re heted nd fibers re drwn from them. he fibers my rndomly bre during the process. he rndom lengths of fibers produced re cut into stndrd lengths immeditely fter their production. ime is divided into periods. Demnd for ech product is ssumed to be nown for ech period nd deliveries re mde t the s of periods. We ssume tht lrge number of preforms re to be processed during ech production period. In the present model we consider only one production period. Our gol is to find such cutting rule tht minimizes the expected loss due to unstisfied demnd.. Rndom Yield he collection of semi-finished products produced from given number of preforms will be clled yield. We ssume tht loctions of microscopic defects in the fiber tht cuse brege form homogenous Poisson process. We chrcterize yield of single preform by the number of semi-finished products of different lengths. Let H totl length of the fiber tht cn be drwn from single preform φ h number of those semi-finished products which lengths fll into intervl [h, h+, where h H hen the rndom yield of single preform is chrcterized by φ = [φ,, φ H ] And the rndom yield of N preforms is ф = φ + + φ N

3. Cutting Rule Let h < <h m be the stndrd product lengths nd h,,h m re integers. Let h be length of semi-finished product, h H. A cutting pttern for the length h is vector such tht v = [v,, v m ] m = v h h, nd v,, v m 0, integer he cutting pttern mens tht we cut v pieces of stndrd length h. Convex combintions of cutting ptterns for length h re clled generlized cutting ptterns. We define cutting rule A s m by H mtrix, where column h is generlized cutting pttern for length h. Suppose the yield ф is cut ccording to A. hen the number of products of length h will be pproximtely ф, where is th row of A. 4. Formultion of the Cutting Rule Finding Problem Let b = [b,, b m ] be demnd for products of length h,,h m. For given cutting rule A nd yield ф [ b Aф ] + is the unstisfied demnd. Let q = [q,, q m ] be cost vector, then the loss due to unstisfied demnd is q [ b Aф ] + Our im is to choose cutting rule tht minimizes the expected loss minimize E[ q [ b Aф ] + ] subject to A is cutting rule

5. Computtion of the Vlue of the Objective Function By multidimensionl limit theorem ф hs H-vrite norml distribution with n expecttion µ nd covrince mtrix C. e b C m C q C b b q A b q E A ] ] [ [ µ π µ µ φ = + + Φ = = 6. Computtion of the Grdient of the Objective Function ΓA is m by H mtrix, nd n entry in th row nd h th column is e b q b q ν σ π ρσ σ ν ν + Φ where µ ν = e h µ ν = h = e h Ce σ C σ = σ σ ρ h Ce = 3

7. Algorithm Fesible Direction Method 0. Initil cutting rule A = 0. Compute ΓA. For ech column h=..h of ΓA solve n integer npsc minimize Γ h A d h subject to m h d h = h h d,..., d m h 0, integer 3. Let D = [ d,, d H ] hen D = D A Is fesible direction tht minimizes the directionl derivtive ΓA 4. Solve one-dimensionl convex problem ' minimize A + λd subject to λ, rel 5. If the length of the step λ is very smll return A ner-optiml cutting rule else A = A + λ D go to 4

8. ests Smll exmple In this exmple the prmeters re: H = 5, m =, probbility q = 0.97, N = 00 demnd b= [60.0, 80.0] costs q= [., 3.4] stndrd lengths: h = h = 4 Solution # IERAIONS = 803 λ = 5.3783e-008 objective vlue = 566.6934 A = [ 0 0.739 0.739 0 0 0 0.6304 0.6304 ] 5

Smll exmple he following prmeters re given in this exmple: H = 5, m =, N = 300, probbility q = 0.97 he stopping tolernce epsilon for lmbd is 0.000 he demnd vector b is [6.0, 37.0] he prices q re [., 3.4] he stndrd fiber lengths h, h re nd 4. Solution # of itertions = 856 λ =.4e-008 objective vlue = 34.8444 A = [ 0 0.3 0.3 0 0 0 0.8389 0.8389 ] 6

9. Mtlb code function [Expecttion,Covrince] = step0h,q,n miu = zerosh,; for h=:h miuh = q^h - logq * q^h * H-h; for h=:h- Expecttionh = miuh - miuh+ * N; ExpecttionH = miuh * N; miu = zerosh,h; for h=:h for g=:h if H-h-g > 0 miuh,g = logq^ * q^h+g * H-h-g^ - * logq* q^g+h * H-h-g + miumxh,g; else miuh,g = miumxh,g; %----------------------------------------------------------------------------------- ex_mh_mg = zerosh,h; ex_mh_mgh,h=miuh; for h=:h- for g=:h- ex_mh_mgh,g = miuh,g - miuh+,g - miuh,g+ + miuh+,g+; %================================================================== Covrince =zerosh,h; for i=:h for =:H Covrincei, = N*ex_mh_mgi, - Expecttioni*Expecttion/N ; 7

function [gmm] = GmmA,q,b,Expecttion,Covrince m = sizea,; H = sizea,; for =:m p = A,:H; if p == 0 for h=:h gmm,h = -q * Expecttionh; else miu = p*expecttion'; sigm_squre = p*covrince*p'; sigm = sqrtsigm_squre; for h=:h miu = Expecttionh; sigm = sqrt Covrinceh,h ; ro = p*covrince:h,h; ro = ro / sigm * sigm ; gmm,h=-q*miu*normcdf b-miu/sigm + q*ro*sigm*/sqrt*pi*exp - /*sigm_squre*b-miu^ ; 8

function [D] = stepha,gmm,hm M = 000; m = sizegmm,; H = sizegmm,; D =zerosm,h; D,=; D,3=; t = [ 0]; t = [ 0]; t3 = [ 0 ]; for h=4:5 obj = Gmm:,h; if t*hm<=h mx = t * obj; x = t; if t*hm<=h o = t * obj; if o < mx mx = o; x = t; if t3*hm<=h o3 = t3 * obj; if o3 < mx mx = o3; x = t3; D:,h = x'; D; D = D - A; function f = Deltx lod'daa', 'A', 'DD', 'q', 'b', 'Ex', 'Cov' m = sizea,; f=0; for =:m = A,:; d = DD,:; if ==0 & d==0 f=f+b*q; else miu = + x * d*ex'; sigm_sure = + x *d *Cov* + x *d '; FF = normcdf b- miu/sqrtsigm_sure ; expon = exp -/*sigm_sure * b - miu ^ ; f = f + q* b- miu * FF + q* sqrtsigm_sure/sqrt*pi * expon; f; 9

function [success] = minq,b,hm,ex,cov,itertions lod'daa_a', 'A' for i=:itertions Gmm = GmmA,q,b,Ex,Cov; DD = stepha,gmm,hm; %step sve'daa', 'A', 'DD', 'q', 'b', 'Ex', 'Cov'; x = fmincon@delt,0.,[],[],[],[],0,; if x<0.000 i x Deltx A A = A + x*dd; return A; A = A + x*dd; i A x Deltx sve'daa_a', 'A' function smll_exmple H = 5; qq = 0.97; N = 300; [Ex,Cov]=step0H,qq,N; hm = [ ; 4 ]; b = [ 6 37 ]; q = [. 3.4 ]; A= [0, 0, 0, 0, 0; 0, 0, 0, 0, 0] sve'daa_a', 'A'; minq,b,hm,ex,cov,3000; 0