Statistical Analysis of Environmental Data - Academic Year Prof. Fernando Sansò

Similar documents
CIVL 7/ D Boundary Value Problems - Quadrilateral Elements (Q8) 1/9

Problem 1: Consider the following stationary data generation process for a random variable y t. e t ~ N(0,1) i.i.d.

CIVL 8/ D Boundary Value Problems - Triangular Elements (T6) 1/8

Consider a system of 2 simultaneous first order linear equations

The Mathematics of Harmonic Oscillators

Math 266, Practice Midterm Exam 2

Introduction to Laplace Transforms October 25, 2017

1 Finite Automata and Regular Expressions

Lecture 4 : Backpropagation Algorithm. Prof. Seul Jung ( Intelligent Systems and Emotional Engineering Laboratory) Chungnam National University

Wave Phenomena Physics 15c

A Hybrid Method to Improve Forecasting Accuracy Utilizing Genetic Algorithm and Its Application to Stock Market Price Data

Generalized Half Linear Canonical Transform And Its Properties

Life Science Journal 2014;11(5s) Evolution of a Helix Curve by observing its velocity

Jonathan Turner Exam 2-10/28/03

Prediction of Aviation Equipment Readiness Rate Based on Exponential Smoothing Method. Yan-ming YANG, Yue TENG and Chao-ran GUO

Advanced Queueing Theory. M/G/1 Queueing Systems

Summary: Solving a Homogeneous System of Two Linear First Order Equations in Two Unknowns

Fourier Series and Parseval s Relation Çağatay Candan Dec. 22, 2013

Chapter 4 A First Analysis of F edback edbac

10.5 Linear Viscoelasticity and the Laplace Transform

INF5820 MT 26 OCT 2012

U1. Transient circuits response

Statistics Assessing Normality Gary W. Oehlert School of Statistics 313B Ford Hall

A Tutorial of The Context Tree Weighting Method: Basic Properties

A parameter robust numerical method for a singularly perturbed Volterra equation in security technologies

Chapter 13 Laplace Transform Analysis

INTERQUARTILE RANGE. I can calculate variabilityinterquartile Range and Mean. Absolute Deviation

3.4 Repeated Roots; Reduction of Order

Title. Author(s)Ito, Yasuhisa; Igarashi, Hajime. CitationIEEE Transactions on Magnetics, 49(5): Issue Date Doc URL. Rights.

Linear System Review. Linear System Review. Descriptions of Linear Systems: 2008 Spring ME854 - GGZ Page 1

SIMEON BALL AND AART BLOKHUIS

Chapter4 Time Domain Analysis of Control System

Major: All Engineering Majors. Authors: Autar Kaw, Luke Snyder

innovations shocks white noise

t=0 t>0: + vr - i dvc Continuation

9. Simple Rules for Monetary Policy

AR(1) Process. The first-order autoregressive process, AR(1) is. where e t is WN(0, σ 2 )

where: u: input y: output x: state vector A, B, C, D are const matrices

Improved Exponential Estimator for Population Variance Using Two Auxiliary Variables

Relation between Fourier Series and Transform

SYMMETRICAL COMPONENTS

Revisiting what you have learned in Advanced Mathematical Analysis

DESIGN OF LOSS FUNCTIONS AND FEATURE TRANSFORMATION FOR MINIMUM CLASSIFICATION ERROR BASED AUTOMATIC SPEECH RECOGNITION MADHAVI VEDULA RATNAGIRI

FL/VAL ~RA1::1. Professor INTERVI of. Professor It Fr recru. sor Social,, first of all, was. Sys SDC? Yes, as a. was a. assumee.

DEPARTMENT OF ELECTRICAL &ELECTRONICS ENGINEERING SIGNALS AND SYSTEMS. Assoc. Prof. Dr. Burak Kelleci. Spring 2018

Laplace Transform. National Chiao Tung University Chun-Jen Tsai 10/19/2011

Control Systems (Lecture note #7)

Performance Implications of Tolerating Cache Faults

Control Systems. Modelling Physical Systems. Assoc.Prof. Haluk Görgün. Gears DC Motors. Lecture #5. Control Systems. 10 March 2013

Preparred by A.Immanuvel Maduram Thangiah, St. John s HSS, Palayamkottai Key for March 2015 Maths Questions Pl.visit 12th-maths-key.weebly.

Supplementary Figure 1. Experiment and simulation with finite qudit. anharmonicity. (a), Experimental data taken after a 60 ns three-tone pulse.

Exclusive Technology Feature. Current-Loop Control In Switching Converters Part 6: Slope Compensation. Slope Compensation. ISSUE: February 2012

Wave Superposition Principle

Let's revisit conditional probability, where the event M is expressed in terms of the random variable. P Ax x x = =

INTEGRAL TRANSFORM METHODS FOR SOLVING FRACTIONAL PDES AND EVALUATION OF CERTAIN INTEGRALS AND SERIES

Improved Exponential Estimator for Population Variance Using Two Auxiliary Variables

PHA Second Exam. Fall 2007

Chapter 5 Transient Analysis

ELEG 413 Lecture #6. Mark Mirotznik, Ph.D. Professor The University of Delaware

State Observer Design

EE Control Systems LECTURE 11

Frequency Response. Response of an LTI System to Eigenfunction

Week 06 Discussion Suppose a discrete random variable X has the following probability distribution: f ( 0 ) = 8

Safety and Reliability of Embedded Systems. (Sicherheit und Zuverlässigkeit eingebetteter Systeme) Stochastic Reliability Analysis

The Variance-Covariance Matrix

The Procedure Abstraction Part II: Symbol Tables and Activation Records

The Equity Index Skew and Asymmetric Normal Mixture GARCH

The Laplace Transform

Canonical Quantizing of Spinor Fields: Anti-Commutation Relations

Conventional Hot-Wire Anemometer

THE ROSENAU-HYMAN K(2,2) EQUATION

EEE 303: Signals and Linear Systems

(A) 1 (B) 1 + (sin 1) (C) 1 (sin 1) (D) (sin 1) 1 (C) and g be the inverse of f. Then the value of g'(0) is. (C) a. dx (a > 0) is

More on FT. Lecture 10 4CT.5 3CT.3-5,7,8. BME 333 Biomedical Signals and Systems - J.Schesser

A general N-dimensional vector consists of N values. They can be arranged as a column or a row and can be real or complex.

Instructors Solution for Assignment 3 Chapter 3: Time Domain Analysis of LTIC Systems

Safety and Reliability of Embedded Systems. (Sicherheit und Zuverlässigkeit eingebetteter Systeme) Stochastic Reliability Analysis

ERAOAL COERECE UCOE CURRE RED ECHOLOGY O 0 l n ll n l Gnlly g l lw hv g l % % xly n g v n n hv g v l h l Bg: R Dg Hgh h x g l lly l lly h ly n HDR n h

Engineering Circuit Analysis 8th Edition Chapter Nine Exercise Solutions

Advanced Engineering Mathematics, K.A. Stroud, Dexter J. Booth Engineering Mathematics, H.K. Dass Higher Engineering Mathematics, Dr. B.S.

Comparison of the performance of best linear unbiased predictors (BLUP)

Let s celebrate! UNIT. 1 Write the town places. 3 Read and match. school. c 1 When s your birthday? Listen, check and practise the dialogues.

Chapter 6 CHOPPER-CONTROLLED DC BRUSH MOTOR DRIVES

16.512, Rocket Propulsion Prof. Manuel Martinez-Sanchez Lecture 3: Ideal Nozzle Fluid Mechanics

The model proposed by Vasicek in 1977 is a yield-based one-factor equilibrium model given by the dynamic

Chapter 1: Review of Quantum Mechanics. Postulates of Quantum Mechanics: 1-3

Final Exam : Solutions

x, x, e are not periodic. Properties of periodic function: 1. For any integer n,

Appendix. In the absence of default risk, the benefit of the tax shield due to debt financing by the firm is 1 C E C

Fourier. Continuous time. Review. with period T, x t. Inverse Fourier F Transform. x t. Transform. j t

Fundamentals of Continuum Mechanics. Seoul National University Graphics & Media Lab

Soft Computing and Energy Time Series

Hybrid Motion Blending Algorithm of 3-Axis SCARA Robot using Parametric Interpolation

Introduction to Inertial Dynamics

Boosting and Ensemble Methods

Chapter 7 Stead St y- ate Errors

A Simple Representation of the Weighted Non-Central Chi-Square Distribution

Ma/CS 6a Class 15: Flows and Bipartite Graphs

- Double consonant - Wordsearch 3

On the Speed of Heat Wave. Mihály Makai

Transcription:

Scl nly of nvronmnl D - cdmc r 8-9 Prof. Frnndo Snò XRISS - PR 5 bl of onn Inroducon... xrc (D mprcl covrnc m)...7 xrc (D mprcl covrnc m)... xrc 3 (D mprcl covrnc m)... xrc 4 (D mprcl covrnc m)...3 xrc 5 (D mprcl covrnc m)...5 xrc 6 (D mprcl vrogrm m)...6 xrc 7 (D collocon)... xrc 8 (D collocon)...3 xrc 9 (D krgng)...4 xrc (D krgng)...7 Upd: 5/6/9 uhor: lbro Moln, L Prun, Mrko Rguzzon

Inroducon In h con h unknown gnl no modlld n drmnc wy (for xmpl lnr combnon of con nd n funcon or lnr combnon of pln funcon) bu n rm of ochc propr (for xmpl mn, covrnc funcon, c.). wo mhod r prnd: collocon nd krgng, h formr umng h h gnl h zro mn, h lr provdng n unbd m whn h gnl h n unknown mn. ollocon um h h obrvon r mpld from rndom proc n m nd hy r compod by gnl plu no: ( ) ( ) ( ),, wh h followng ochc fur: [ ( )] [ ( )] [ ( ) ( )] ( ), gnl covrnc funcon [ ( ) ( )] ( ) δ [ ( ) ( )],, no covrnc funcon, wh δ mnng h gnl nd no r uncorrld. W wn o m h gnl vlu n by ung lnr combnon of h vlbl obrvon: ( ). In ordr o drmn h combnon wgh, w nvok h Wnr-Kolmogorov prncpl: {[ ( ) ( )] } nmly w wn o mnmz h mn qur mon rror. h rul of h mnmzon : whr: (, ) (, ) K (, ) mn ( ) ( ) [x] [x] [x] (, ) (, ) K (, ) (, ) (, ) K (, ) M O (, )... (, ) M O M K K I wh h numbr of vlbl obrvon.

If n obrvon pon w lk bou flrng, bcu w wn o pr h gnl from no; f no n obrvon pon w lk bou prdcon. h mon rror h dffrnc bwn h ru gnl nd h md gnl, nmly: ( ) ( ) ( ). w do no know h ru gnl, w cn only m h rror vrnc: ( ) ( ) { } ( ) ( ),. L u conclud wh condron on h gnl covrnc mrx rucur. In gnrl, ymmrc mrx follow: ( ) ( ) ( ) ( ) ( ) O O K K 3 3,,,,, umng h h d r xrcd from onry proc n m: () [ ] con, h vrfd bcu ( ) [ ] ( ) ( ) [ ] ( ) ( ), h rulng gnl covrnc mrx : ( ) ( ) ( ) ( ) ( ) O K K 3 3 Fnlly, um h h d r rgulrly drbud n m (grd pon): w g h followng gnl covrnc mrx: 3 4 Δ ( ) ( ) ( ) ( ) ( ) ( ) Δ Δ Δ h o-clld oplz mrx. 3

Krgng ow um h h obrvon r mpld from rndom proc wh n unknown mn vlu h n gnrl dffrn from zro. h obrvon cn b modlld : whr: [ ( )] [ ( )] μ [ ( )] ( ) μ ( ) ( ) ( ) μ u( ) ( ) u unknown mn μ u u, [( )( ( ) )] [ ( ) ( )] u ( ) [ ( ) ( )] u (, ) μ [ ( ) ( )] ( ) [ ( ) ( )] [ u( ) ( )], gnl covrnc funcon, no covrnc funcon mnng h gnl nd no r uncorrld. Snc μ unknown, h drc m of h gnl covrnc funcon from h d qu complcd; hrfor h followng funcon dfnd n uch wy h do no dpnd on h mn vlu: [ ] [ ( u( ) u( ) ] ( ) (, ) ( ( ) ( ) u, h funcon clld vrogrm. W wn o m h gnl vlu n by ung lnr combnon of h vlbl obrvon: nd dmndng h h m unbd: h condon corrpond o forc h: ( ) [ ( )] μ.. whch n gnrl no fd n h c of collocon. In ordr o drmn h combnon wgh, h followng ym h o b olvd: ( Γ ) whr: α Γ M Γ (, ) (, ) M (, ) Γ M (, ) (, ) K (, ) (, ) (, ) K (, ) (, ) K K (, ) O M 4

I O wh h numbr of vlbl obrvon nd α h Lgrng mulplcor. h vrnc of h mon rror ( ) ( ) ( ) gvn by: ( ) [ ( )] α Γ. o h h vrnc of h mon rror of h krgng oluon lwy lrgr (or mo qul) hn h corrpondng vrnc of h collocon oluon. On h ohr hnd h unbn condon no gnrlly fd by collocon. ovrnc funcon mon h procdur con of hr p: mprcl covrnc funcon mon umng h h D proc onry n m (nvrn by rnlon n m) or, n gnrl, h h nd rndom fld homognou nd oropc (nvrn by roo-rnlon): mp ( ) covrnc funcon nrpolon ung pov dfn modl, lk for xmpl: ( ) ( ) ( ),, ( ) > whr, r prmr o b md. o h h vlu of h mprcl covrnc funcon n h orgn hould no b ud n h gnl covrnc funcon nrpolon bcu h um of h gnl nd no vrnc. no vrnc mon: ( ) ( ) ( ) ( ) mp If > mp for numrcl ron, hn h vrnc of h no h o b forcd qul o zro or up o n -pror vlu. 5

mp( ) Ĉ ( ) Vrogrm mon I lk h covrnc funcon mon, bu now h mprcl vrogrm vlud from h d. hr rlon bwn covrnc funcon nd vrogrm: ( ) ( ) ( ) ( ) h clld nugg ffc Howvr hr x vrogrm modl h do no hv h corrpondng covrnc modl, lk for xmpl: ( ) < ondr for xmpl, h covrnc modl no ccpbl bcu h condon no fd. ( ) ( ) ( ) ( ) < R ( ) ( ) ( ) ( ) no ccpbl 6

xrc ondr h followng obrvon: 3 4 5 6 7 8 9.6446.7845 -.646 4.84 5..7479 -.9-4.4.393-5.8349 m h mprcl covrnc funcon for 4, umng h h proc onry nd zro mn. Inrpol h mprcl covrnc funcon wh n xponnl modl: ( ). m h ndrd dvon of h wh no. Fr of ll, compu h mprcl covrnc funcon: mp mp mp ( ). 46 mp () 3 () ( ), 3. 47,. 6436 mp( ) mp 3 3 ( ) 4 4 4, 3 -.94, 4 -.479 In ordr o m h gnl vrnc, whch corrpond o h prmr of h covrnc nd w xrpol h vlu n : funcon, w connc wh rgh ln ( ) nd ( ) mp ( ) mp mp quon of h rgh ln png hrough wo pon: mp () ( ) mp mp mp ( ) ( ) ( ) mp 3.5858 7

.635 mp ( ) In ordr o g fr m of h prmr, w compu h corrlon lngh, h h m dnc for whch h covrnc funcon qul o hlf h vrnc:. 799 corrlon lngh c hn l u forc h h nrpold covrnc funcon qul o h mprcl covrnc funcon for : c [ ]. 54 ( ) mp log logmp ( ) Summrzng, fr m of h covrnc modl prmr 3.5858 nd. 54, whl h no ndrd dvon rul υ.635. In ordr o mprov h rough m, (lnrzd) l qur dumn cn b up. h fr p o lnrz h xpron of h covrnc modl wh rpc o h unknown prmr: whr: ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) hn clcl l qur problm cn b olvd, condrng ll h vlbl obrvon (corrpondng n h c o h mprcl covrnc funcon), pr from : () ( ) () 3 ( 4) mp 3.47 mp.6436 obrvon mp.94.479 mp.8586-3.788.737-5.87 3 3 3 dgn mrx.633-5.8468 4 4 4.5435-7.7957 (don mx up wh h covrnc modl prmr ) mp ( ) 8

.9489.698.6436 3.788 4 3 known rm (don mx up wh h covrnc modl prmr ) I Q x δ δ δ norml Mrx 3.394-4.4796-4.4796.9769 ( ).753 3.33 x δ.54 3.5858 x.947 6.988 x x x δ ( ) 878. mp o h h l qur oluon hould b rd by condrng nw pproxmon vlu h m of h prvou p, unl h convrgnc rchd. 9

xrc ondr h m obrvon xrc. um h h d r mpld from onry proc wh n unknown mn. m h mn of h proc from h d. Rmov h md mn vlu from h d. m h mprcl covrnc funcon from h rdul d, for 4. Inrpol h covrnc funcon wh n xponnl modl: ( ). m h ndrd dvon of h wh no. Fr of ll, l u m h mn of h proc : m o h w cn obn h followng rducd obrvon:.368 3 4 5 6 7 8 9 m.478.5477 -.54 4.5673 4.8634.5 -.39-4.573.75-6.77 ompu h mprcl covrnc funcon for 4: mp( ) mp mp (). 3599 ( ). 9358,. 4677, mp mp () 3 3 ( ) 4 4 Drv fr m of h covrnc modl prmr, n xrc. quon of h rgh ln png hrough wo pon: mp () ( ) mp mp 3.439.6374 mp ( ) mp ( ) ( ) ( ) mp 3 -.4453 4, 3 -.46, 4

In ordr o g fr m of h prmr, w compu h corrlon lngh, h h m dnc for whch h covrnc funcon qul o hlf h vrnc: 7. corrlon lngh c hn l u forc h h nrpold covrnc funcon qul o h mprcl covrnc funcon for c : ( ) ( ) 68. log log mp mp Summrzng, fr m of h covrnc modl prmr 439 3. nd 68., whl h no ndrd dvon rul 6374. υ. Improv h m of h prmr by pplyng l-qur dumn, n xrc : () ( ) () ( ).46.4453.4677.9358 4 3 mp mp mp mp 7.56 -.556-5.367.673 4.9355 -.75.8983 -.855 4 3 4 4 3 3.789..4677.8983 4 3 I Q x δ δ δ.7747-3.99-3.99.978 ( ).849 3.54 x δ.68 3.439 x.7 6.943 x x x δ ( ) 8589. mp o h h l qur oluon hould b rd by condrng nw pproxmon vlu h m of h prvou p, unl h convrgnc rchd.

xrc 3 ondr h followng obrvon: 3 4 5 6 7 8 5.75.788.58 -.789 -.383 -.46-3.86 -.9 m h mprcl covrnc funcon lg, lg, lg, umng h h proc onry nd zro mn. m h gnl covrnc funcon uppong h blong o h fmly: ( ). m h ndrd dvon of h wh no. Fr of ll, compu h mprcl covrnc funcon: mp mp 4 mp, ( ) 8. ( ), 5. 867 () mp In ordr o m h covrnc modl prmr, l u wr h ym: mp () ( ) 3 mp mp ( ) ( ) 3. 5 ( ) ( ) 5.867 mp log log.938 3 mp 3 3.5 By ung h fr quon: () 5.867.938 mp mp 7.67 ( ) 8.4-7.67. 664

xrc 4 ondr h followng obrvon n h pln: P P P 3 P 4 P 5 P 6 P 7 x...3.5 4.3 3.9.4 x.3 4.7.5 3.5.8 3.. -.44 -.84.4.3 -.5.9.7 m h mprcl covrnc funcon umng h h rndom fld zro mn nd homognou nd oropc. ompu h mprcl covrnc funcon for,.5,. 5,. 5. Inrpol h mprcl covrnc funcon wh n xponnl modl: ρ ( ρ ) m h no vrnc. ( x x ) ( x x ) ρ. Fr of ll, l u rprn h obrvon pon: x hn, compu h mrx of h dnc bwn ch coupl of pon (ymmrc mrx wh zro dgonl): D ( ).64.846.969.447.387..3.96.4.8544.4.597.7464.86.5657 W cn now compu h mprcl covrnc funcon: mp.79 3.557.4698.58.63.66 ( ).56 7 3 4 5 6 7 3 4 5.34 5 (.5) ( ). 68 mp 3 4 4 5 4 6 5 6, d (,] x 3

9 34 36 47 57 5 6 7 5 6 (.5). 53 mp, (,] d 6 (.5) ( ). 98 mp 3 4 3 5 3 7 6 7 3, d (,3] W cn now compu fr m of h covrnc modl prmr by forcng h h modl p hrough (. 5) nd (. 5). W g h followng ym: mp (.5) (.5) mp (.5) (.5).5 mp.68 mp 4.9866.5.53 mp mp log 4.9866. 668 ondrng h fr quon, w obn:.668.5.68.68. 584.4478 h no vrnc rul:.56.584. mp ( ) 469 fr lnrzng h xpron of h covrnc modl: ( ρ ) ( ρ) ( ρ) ( ) ( ) ρ ( ) w cn mprov h m of h modl prmr by pplyng l-qur dumn:.5.5 (.5).68.5 mp.4478 -.34 ( ) ( ) mp.5.53.5.5.5.898 -.784.5.98.5.5 mp.5.8 -.6.5.68.5 δ.53 Q I δ x.5.5 δ.89 -.659 -.659.38 ( ) δ x.4.46.584 x.58 x x δ x.668.6 ( ).56.58. 476. 686 mp In h c h corrcon from l qur dumn vry mll. 4

xrc 5 ondr h followng obrvon n h pln: m h mprcl covrnc funcon umng h h rndom fld zro mn nd homognou nd oropc. ompu h mprcl covrnc funcon for,. 5,. 75. Inrpol h mprcl covrnc funcon wh n xponnl modl: ( ρ ) ρ ρ ( x ) ( ) x x x m h wh no vrnc. P P P 3 P 4 P 5 P 6 x.3.3.9.7. -. x.9..3.7.3.4-3.89 4.88 5.9.3 3.95.99 fr rprnng h obrvon pon: x compu h mrx of h dnc bwn ch coupl of pon: D ( ).7.8485.683.447.643.447.635.36.8.7 W cn now compu h mprcl covrnc funcon: mp.583.366.955.766.44 ( ) 4. 659 6 3 4 5 6 5 (.5) ( ) 6. 5 mp 4 5 6 3 4 5 6, (,.5] mp d.5.5.75 3 5 6 3 (.75). 563, 4 35 36 45 46 d (.5,] x 5

W cn now compu fr m of h covrnc modl prmr by forcng h h modl p hrough (. 5) nd (. 5). W g h followng ym: mp.5.75 6.5.563 mp.5 4. 899 log 4.899. 87 Ind of ubung h md n on of h wo quon nd hn drv h corrpondng, w wr mpl l qur problm: B.5.75.87.5.87.75 6.5 obrvon vcor.563 B B B 7. h no vrnc rul: ( ) 395 4.659 7.395 7.38.689.8386 dgn mrx.5 xrc 6 ondr h followng obrvon, rrgulrly drbud n m:.84.86.74 3.87 4.76 6.4 7. 8.3 9.35.4.89 5.53 6.8 7.85 8. 3.7 9.3 9.95..7 3.6 3.89 4.96 6.8 36.7.57 3.4 9.99.35 8.4 3.3 6.4 m h mprcl vrogrm for < 8. vrg h mprcl vrogrm ovr nrvl of Δ. Inrpol h mprcl vrogrm (h orgnl on nd h vrgd on) wh lnr modl:. m h wh no vrnc. ( ) b 6

h obrvon cn b modlld : whr: [ () ] [ () ] u [ () ( ' )] δ ' ( ) μ u( ) ( ) [( u() u( ' )) ] ( ' ) ( ) Fr of ll, compu h mrx of h dnc bwn ch coupl of pon: D..9.88 3.3..3 3.9 5.4.9 4.38. 3.5.89.37.48 6.6 5.4 4.6 3.3.4.76 7.9 6.7 5.9 4.6 3.7.79.3 8.48 7.46 6.58 5.45 4.56 3.8.3.9 9. 8.9 7. 6.8 5.9 3.7.95.9.63.37 9.35 8.47 7.34 6.45 4.97 4. 3.8.89.6.88 9.86 8.98 7.85 6.96 5.48 4.7 3.69.4.77.5.3.3.4 9.9 8.4 6.9 6.6 5.3 3.84 3..95.44 3.5.3.5. 9.3 7.65 6.89 5.86 4.57 3.94.68.7.73 4. 3...9. 8.7 7.96 6.93 5.64 5. 3.75 3.4.8.7 5.34 4.3 3.44.3.4 9.94 9.8 8.5 6.86 6.3 4.97 4.46 3..9. whr d. For ch dnc vlu, w cn compu h corrpondng mprcl vrogrm vlu, obnng h followng mrx: Γ ( mp) whr ( mp) 59.96 66.5858.3 3.896 3.99.7648 [ ] ( ) ( ( ) ( ) 5.64.648 7.584.83 mp nd. 7.5 8.956 4.8.69.535 O L K K K K K 7

cr plo gnrlly ud o rprn h cloud of h mprcl vrogrm vlu: ( mp ) () h mprcl vrogrm h o b nrpold wh propr modl; o h m uful o vrg, o o dnfy h hp of h vrogrm modl. L u condr Δ, w g: ( mp ) ( ) ( mp ) ( mp) () [ 59.96 66.5858.... ] 9. 494 < d 6 ( mp ) ( mp) () 3 [ 3.896 5.64 3.99... ] 35. 364 3 d 4 < ( mp ) ( mp) () 56.63 ( ) 5 5 5 4< d 6 ( mp ) ( mp) ( ) 7.5649 ( 9) 7 7 7 6< d 8 Rprnng h vrgd mprcl vrogrm, y o rcognz lnr rnd: (mp) ( ) 4 6 8 ow w cn compu h prmr of h vrogrm modl: ( ) b > 3 5 7 by pplyng l qur dumn o h orgnl (no vrgd) mprcl vrogrm, nglcng h vlu n h orgn whch lwy qul o zro: 8

59.96 66.5858 3.896 5.64 M..9 3.3 3.9 M M Q I 687.4 336. 336. 87 numbr of vlu of h mprcl vrogrm for < 8.57.994.994.4988 9.58.55 x 3856.37 b 3.563 ( ) mp ( ) h no vrnc rul: b 3.563. b 4 6 8 h modl nrpolon cn b lo compud from h vrgd vrogrm, hu rducng h compuonl burdn, bu nroducng om pproxmon du o h dffrn ccurcy of h md mprcl vrogrm vlu (h hould b modlld n h Q mrx of h l-qur problm). 9.494 35.364 56.63 7.5649 3 5 7 Q I 84 6 6 4.5...5 96.35 9.34 x 83.963 b 9.86 9

(mp) () h no vrnc rul: b.897. b 3 5 7

xrc 7 ondr wo obrvon ( fgur) h r xrcd from D onry proc wh zro mn nd known covrnc funcon ( ). h obrvon r ffcd by wh no wh known vrnc.5. 3.837.35 - m h gnl vlu n h mn pon bwn h wo obrvd vlu ( ), ccordng o h Wnr Kolmogorov prncpl. ompu h vrnc of h mon rror. h obrvon cn b modlld : whr [] [ ] [ ] (gnl nd no r uncorrld) In ordr o olv h collocon problm, w compu h gnl covrnc mrx n h obrvon pon: ( ) ( ) ( ) ( ) h no covrnc mrx n h obrvon pon: I nd h gnl covrnc vcor bwn h obrvon pon nd h prdcon pon: () () h collocon m of h gnl n rul: ( ) ( )

whr:.353.5.353.5.5.5.353.353.5.353.353.5.353.895.5.876.895.876.876.876.895 [.3679.3679] [.656. 656].35 3.837.895 ( ) [.656.656]. 34 o h bcu h wo obrvon hv h m no vrnc nd hv h m dnc o h prdcon pon. o lo h, hppn nd wh krgng. h vrnc of h mon rror rul: ( ) ( ) [.3679.3679] [.656.656].954. 846.895.876.3679.3679.876.3679.895.3679 If w do no um h h proc zro mn, fr w hv o m h mn of h proc by ung h covrnc nformon o dfn h Q mrx of h l qur problm: whr: μ.35 3.837 ( Q ) Q [.5.5].544 m μ Q.35 3.837 hn w cn pply collocon o h rdul d (fr ubrcng h md mn) nd, n h nd, w cn ror h rmovd mn vlu o h gnl m:.893 μ.893 ( ) ( μ) [.656.656].544.544.544 mo h procdur h o b ppld only whn h proc no umd o b zro mn. If h covrnc funcon known, h procdur ld o h m rul of krgng.

xrc 8 ondr h followng obrvon n h pln: x x x P.6 P -.883 P -.9378 P P P P x m h gnl vlu n (,) P umng h h obrvon r mpld from homognou nd oropc rndom fld wh zro mn nd covrnc funcon ρ 4 ( ρ) ρ (wh ρ h ucldn dnc bwn coupl of pon). h obrvon r corrupd by wh no wh vrnc. 5 ompu h vrnc of h mon rror. ( ρ ). ρ h obrvon cn b modlld : whr: (,) ( ), ( ), ( ) ( ) ( ) ( ).3679 ( ) ( ) ( ) ( ) ( ) 3.3679.46 3.3679.46.3679.5.5.5 h opml combnon wgh ccordng o h Wnr Kolmogorov prncpl r: ( ) [. 78.7767.7767 ] 3

whr: ( ).46 ( ).3679 ( ).3679 ( ).76.469.469.469.988.4949.469.4949.988 h collocon m n P (, ) rul: ( P ) ( ). 776 wh h followng mon rror vrnc: ( P ) ( ) ( ) ( ).889. xrc 9 ondr h m obrvon xrc 7, now umng h h proc h n unknown conn mn. m h gnl vlu n nd h corrpondng mon rror vrnc by pplyng krgng. 3.837 ( ) (gnl covrnc).35.5 (wh no vrnc) - h obrvon cn b modlld : whr [] μ [] [ ] (gnl nd no r uncorrld) I 4

W wn o m h gnl n lnr combnon of h vlbl obrvon: ( ) whr h coffcn hv o b drmnd by krgng. Fr of ll, drv h vrogrm from h covrnc funcon: ( ) ( ) ( ) ( ) () () ( ) ( ) ( ) hn w compu h vrogrm mrx n h obrvon pon: ( ) Γ nd h vrogrm vcor bwn h obrvon pon nd h prdcon pon: ( ) whr: 5

h krgng ym cn b wrn : ( ) Γ α whr: Γ By xpndng, w g ym of hr quon wh hr unknown (h numbr of unknown of krgng oluon lwy qul o h numbr of obrvon plu on). ( ) ( ) α α h oluon of h ym : M 444 4 3 4 44 4 α ( ) ( ) 93. d M ( ) ( ).373.5.5.5.4486.4486.5.4486.4486 d M M.348.5.5.63.63 M α hrfor, h krgng m of h gnl n rul: ( ) 544. 3.837.35 h vrnc of h mon rror gvn by: ( ) ( ).9569.63.348.63.63.348 Γ α o h h krgng m corrpond o h mn vlu of h wo vlbl obrvon. o lo h h mon rror vrnc of h krgng oluon hghr hn h corrpondng vrnc of h collocon oluon, xpcd. 6

xrc ondr h m obrvon xrc 8, now umng h h proc h n unknown conn mn. m h gnl vlu n P (,) nd h corrpondng mon rror vrnc by pplyng krgng. x x P.6 P -.883 P -.9378 P x P P P x h obrvon cn b modlld : whr now [] μ. h vrogrm drvd from h gvn covrnc funcon ρ 4 ( ρ) ρ ( ) ( ρ) ρ 4 Γ ρ In ordr o olv h krgng problm w compu h followng mrc: 3 3.5.5.5 3 h krgng ym cn b wrn : whr: Γ α M Γ.5.3679.3679.3679.5.46.3679.46.5 7

.46.3679 b.3679 h oluon of h ym : M 3 α.3.3439 b.3439.5433 o h h krgng m of h gnl rul: ( P ).3.6.3439 (.883).3439 (.9378). 639 h mon rror vrnc gvn by: ( P ) α.5433.3758. 93 whch hghr hn h on obnd by collocon. 8