Monte Carlo Simulation Monte Carlo Simulation - University of Florida

18
Monte Carlo Simulation A7G6VQPPN2A.jpg How do we generate the statistics of a function of a random Monte Carlo Simulation pics.hoobly.com/full/AA How do we generate the statistics of a function of a random variable? Why is the method called “Monte Carlo?” SOURCE: http://p How do we use the uniform random number generator to generate other distributions? A th di t ib ti di tl il bl i tl b? Are other distributions directly available in matlab? How do we accelerate the brute force approach? Probability distributions and moments Probability distributions and moments Web links: htt // ikl /li k/ t l th d ht http://www.riskglossary.com/link/monte_carlo_method.htm http://physics.gac.edu/~huber/envision/instruct/montecar.htm

Transcript of Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Page 1: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Monte Carlo Simulation A7G

6VQ

PP

N2A

.jpg

How do we generate the statistics of a function of a random

Monte Carlo Simulation

pics

.hoo

bly.

com

/full/

AA

How do we generate the statistics of a function of a random variable?– Why is the method called “Monte Carlo?” S

OU

RC

E: h

ttp://

p

How do we use the uniform random number generator to generate other distributions?

A th di t ib ti di tl il bl i tl b?– Are other distributions directly available in matlab?

How do we accelerate the brute force approach?Probability distributions and moments– Probability distributions and moments

Web links: htt // i k l /li k/ t l th d hthttp://www.riskglossary.com/link/monte_carlo_method.htmhttp://physics.gac.edu/~huber/envision/instruct/montecar.htm

Page 2: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Basic Monte Carlo

iles/

0410

2210

4413

s.jp

g

• Given a random variable X and a function h(X):

E: w.s

z-w

hole

sale

.com

/upl

oadF

i

sample X: [x1,x2,…,xn];Calculate [h(x1),h(x2),…,h(xn)]; use to approximate statistics of h.

SO

UR

CE

http

://w

ww

• Example: X is U[0,1]. Use MCS to find mean of X2

x=rand(10); y=x.^2; %generates 10x10 random matrix( ) y gsumy=sum(y)/10sumy =0.4017 0.5279 0.1367 0.3501 0.3072 0.3362 0.3855 0.3646 0.5033 0.2666sum(sumy)/10 ans =0 3580sum(sumy)/10 ans =0.3580

• What is the true mean

SOURCE: http://schools.sd68.bc.ca/ed611/akerley/question.jpg

Page 3: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Basic Monte Carlo• Given a random variable X and a function h(X): sample

X: [x1,x2,…,xn]; Calculate [h(x1),h(x2),…,h(xn)]; use to [ 1, 2, , n]; [ ( 1), ( 2), , ( n)];approximate statistics of h.

• Example: X is U[0,1]. Use MCS to find mean of X2

x=rand(10); y=x.^2; %generates 10x10 random matrixsumy=sum(y)/10sumy =0 4017 0 5279 0 1367 0 3501 0 3072sumy =0.4017 0.5279 0.1367 0.3501 0.3072

0.3362 0.3855 0.3646 0.5033 0.2666sum(sumy)/10 ans =0.3580( y)• What is the true mean SOURCE:

http://schools.sd68.bc.ca/ed611/akerley/question.jpg

Page 4: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Obtaining distributionsObtaining distributions

• Histogram: y=randn(100 1); hist(y)Histogram: y=randn(100,1); hist(y)25

15

20

10

15

0

5

-3 -2 -1 0 1 2 30

Page 5: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Cumulative density functionCumulative density function• Cdfplot(y)p (y)

0 8

0.9

1Empirical CDF

0.6

0.7

0.8

)

0.3

0.4

0.5

F(x

-3 -2 -1 0 1 2 3 40

0.1

0.2

3 2 1 0 1 2 3 4x

Page 6: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Experimental CDFExperimental CDF

• [f x]=ecdf(y); f’[f,x]=ecdf(y); fColumns 1 through 8

0 0.0100 0.0200 0.0300 0.0400 0.0500 0.0600 0.0700Columns 9 through 16Columns 9 through 16

0.0800 0.0900 0.1000 0.1100 0.1200 0.1300 0.1400 0.1500

• x‘Columns 1 through 8

-2.2529 -2.2529 -2.1763 -2.1746 -1.9539 -1.9068 -1.6354 -1.5729Columns 9 through 16

1 4127 1 2568 1 2203 1 2158 1 1957 1 1684 1 1645 1 0954-1.4127 -1.2568 -1.2203 -1.2158 -1.1957 -1.1684 -1.1645 -1.0954

Columns 89 through 960.9940 1.0437 1.0945 1.1654 1.2020 1.2750 1.5081 1.5338

Columns 97 through 1011.5908 1.7909 1.8918 2.4124 2.7335

Page 7: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Histogram of averageg g• x=rand(100); y=sum(x)/100; hist(y)

30

35

20

25

15

20

5

10

0.42 0.44 0.46 0.48 0.5 0.52 0.54 0.56 0.580

Page 8: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Histogram of average• x=rand(1000); y=sum(x)/1000; hist(y)

140

100

120

80

100

40

60

0 46 0 47 0 48 0 49 0 5 0 51 0 52 0 530

20

0.46 0.47 0.48 0.49 0.5 0.51 0.52 0.53

What is the law of large numbers?

Page 9: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Distribution of x2Distribution of x

x=rand(10000,1);( , );x2=x.^2;hist(x2,20)

2500

1500

2000

1000

1500

500

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

Page 10: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Other distributions• Other distributions available in matlab• For example, Weibul distributionFor example, Weibul distribution

/1( )bx ab bf x ba x e

600

( )0

f x ba x efor x

400

500

300r=wblrnd(1,1,1000);hist(r,20)

100

200

0 2 4 6 8 10 12 140

Page 11: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Correlated Variables

For normal distrib tion can se Matlab’s m nrnd• For normal distribution can use Matlab’s mvnrnd• R = MVNRND(MU,SIGMA) returns an n-by-d

matrix R of random vectors chosen from thematrix R of random vectors chosen from the multivariate normal distribution with mean vector MU, and covariance matrix SIGMA. MU is anMU, and covariance matrix SIGMA. MU is an n-by-d matrix, and MVNRND generates each row of R using the corresponding row of MU. SIGMA is a d-by-d symmetric positive semi-definite matrix

Page 12: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Examplepmu = [2 3];sigma = [1 1 5; 1 5 3];sigma = [1 1.5; 1.5 3];r = mvnrnd(mu,sigma,20);plot(r(:,1),r(:,2),'+') 4.5

5

plot(r(:,1),r(:,2), )

3.5

4

2

2.5

3

1

1.5

0 0.5 1 1.5 2 2.5 3 3.50

0.5

Page 13: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Latin hypercube samplingyp p g

X = lhsnorm(mu,SIGMA,n) generates a latinX lhsnorm(mu,SIGMA,n) generates a latin hypercube sample X of size n from the multivariate normal distribution with mean vector mu and covariance matrix SIGMA. X is similar to a random sample from the

lti i t l di t ib ti b t thmultivariate normal distribution, but the marginal distribution of each column is adjusted so that its sample marginaladjusted so that its sample marginal distribution is close to its theoretical normal distributionnormal distribution.

Page 14: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Comparing MCS to LHS

mu = [2 2];sigma = [1 0; 0 3];sigma [1 0; 0 3];r = lhsnorm(mu,sigma,20);sum(r)/20

r = mvnrnd(mu,sigma,20);sum(r)/20

2 3327 2 2184( )

ans = 1.9732 2.0259

5

6

4

5

ans =2.3327 2.2184

2

3

4

2

3

-1

0

1

-1

0

1

0.5 1 1.5 2 2.5 3 3.5 4 4.5-2

0 0.5 1 1.5 2 2.5 3 3.5 4-2

Page 15: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Evaluating probabilities of failureEvaluating probabilities of failure

• Failure is defined in terms of a limit stateFailure is defined in terms of a limit state function that must satisfy g(r)>0, where r is a vector of random variables. /fP m N

• Probability of failure is estimated as the ratio of number of negative g’s, m, to total MC sample

f

size, N• The accuracy of the estimate is poor unless N

i h l h 1/Pis much larger than 1/Pf

• For small Pf

(1 )

1

f ff

f

P PP

NP

1f

f

PP m

Page 16: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Separable Monte CarloSeparable Monte Carlo

• Usually limit state function is written inUsually limit state function is written in terms of response vs. capacity g=C(r)-R(r)>0g C(r) R(r)>0

• Failure typically corresponds to structures with extremely low capacity or extremelywith extremely low capacity or extremely high response but not bothC t k d t f th t i bl• Can take advantage of that in separable MC

Page 17: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

Reading assignmentg gSDM paper by Ben Smarslok

Source: www.library.veryhelpful.co.uk/ Page11.htm

Page 18: Monte Carlo Simulation Monte Carlo Simulation - University of Florida

problems1. Derive formula for the standard

deviation of estimate of Pfdeviation of estimate of Pf

2. Use MCS to calculate the probability of failure when astructure is designed with 50% safety factor

Source: Smithsonian InstitutionNumber: 2004-57325

structure is designed with 50% safety factor and both stress and failure stress have a normal distribution with 10% COV.normal distribution with 10% COV.

3. Calculate the exact value of the answer to Problem 2 (that is without MCS).