System Simulation Part II: Mathematical and Statistical Models Chapter 5: Statistical Models

Transcription

1 System Simulation Part II: Mathematical and Statistical Models Chapter 5: Statistical Models Fatih Cavdur March 29, 2014 Introduction Introduction The world of the model-builder sees probabilistic rather than deterministic. There are many causes of variation. The time it takes to fi a broken machine is a function of the compleity of the breakdown, whether the repairman brought the proper replacement parts and tools, whether the operator receives a lesson in preventive maintenance etc. To the model-builder, these occur by chance and cannot be predicted, however, some statistical model might well describe the time to make a repair.

2 Review of Terminology and Concepts Discrete Random Variables If the number of possible values of X is finite or countably infinite, X is called a discrete RV. The possible values of X are given by the range space of X, denoted by R X. Here, R X = 0, 1,.... Let X be a discrete RV. With each possible outcome i in R X, a number, p( i ) = P(X = i ) gives the probability that the RV equals the value of i. We have p( i ) 0 for all i i=1 p( i) = 1 p( i ) is called the probability mass function (PMF) of X. Review of Terminology and Concepts Discrete Random Variables-cont. Consider the eperiment of tossing a single biased die. Define X as the number of spots on the up face, R X = 1, 2, 3, 4, 5, 6. Here, the probability that a given face lands up is proportional to the number of spots showing. So, the discrete probability distribution of of this eperiment is given by i p( i ) 1/21 2/21 3/21 4/21 5/21 6/21

3 Review of Terminology and Concepts Discrete Random Variables-cont. p() 6/21 5/21 4/21 3/21 2/21 1/ Figure : PMF for the Die Eample Review of Terminology and Concepts Continuous Random Variables Continuous Random Variables (Continuous RV): If the range space R X of the RV X is an interval or a collection of intervals, X is called a continuous RV. We have P(a X b) = We also have f () 0 for all in R X R X f ()d = 1 f () = 0 if is not in R X b a f ()d

4 Review of Terminology and Concepts Continuous Random Variables-cont. f () is called the probability density function (PDF) of the RV X. Also, P(X = 0 ) = and, 0 0 f ()d = 0 P(a X b) = P(a < X b) = P(a X < b) = P(a < X < b) Review of Terminology and Concepts Continuous Random Variables-cont. f() a b Figure : Graphical Interpretation of P(a < X < b)

5 Review of Terminology and Concepts Continuous Random Variables-cont. The life of a device used to inspect cracks in aircraft wings is given by X, a continuous RV with a PDF of { 1 f () = 2 e /2, 0 The probability that the life of the device is between 2 and 3 years is P(2 X 3) = e /2 d = e 3/2 + e 1 = Review of Terminology and Concepts Continuous Random Variables-cont. f() Figure : PDF for the Eample Problem

6 Review of Terminology and Concepts Cumulative Distribution Function The cumulative distribution function (CDF), denoted by F (), shows the probability that F () = P(X ). If X is discrete, f () = i p( i ) If X is continuous, F () = f (t)dt Review of Terminology and Concepts Cumulative Distribution Function-cont. Some properties of the CDF are F is a non-decreasing function. lim F () = 0 lim F () = 1 All probability questions about X can be answered in terms of the CDF, such as P(a < X b) = F (b) F (a), for all a < b Look at the eamples in your tet.

7 Review of Terminology and Concepts Epectation and Variance Epectation of X if X is discrete, E(X ) = i p( i ) i Epectation of X if X is continuous, E(X ) = + f ()d Variance of X (σ 2, Var(X ), V (X )) V (X ) = E ( [X E(X )] 2) = E (X ) 2 [E(X )] 2 Useful Statistical Models Probability Distributions Consider the single-server-queuing system. The arrivals and service times might be deterministic or probabilistic. We can use probability distributions to model the stochastic features. For eample, for service times if the data are completely random, the eponential distribution might be considered if the data fluctuate from some value, the normal or the truncated normal distribution might be used the gamma and Weibull distributions can also be used to model inter-arrival and service times Some others, such as the geometric, Poisson and negative binomial distributions might be considered for demand distribution etc.

8 Discrete Distributions Bernoulli Distribution Consider an eperiment with n trials each of which can be a success of failure with the corresponding probabilities. It is called a Bernoulli process if the trials are independent and the probability of success is the same for all trials. We have p j ( j ) = p( j ) = p, j = 1, j = 1, 2,..., n 1 p = q, j = 0, j = 1, 2,..., n E(X j ) = 0 q + 1 p = p V (X j ) = 0 2 q p p 2 = pq Discrete Distributions Binomial Distribution PDF p() = { ( n ) p q n, = 0, 1, 2,..., n Epectation E(X ) = np Variance V (X ) = npq

9 Discrete Distributions Binomial Distribution-Eample A production process manufactures computer chips that are defective 2% of the time on average. Every day a random sample of size 50 is taken from the process. If the sample contains more than 2 defectives, the process is stopped. Compute the probability of the process stopping. Discrete Distributions Binomial Distribution-Eample If we let X be the number of defectives in the sample, then, X have a binomial distribution with parameters (n = 50, p = 0.02). Hence, P(X > 2) = 1 P(X 2) = 2 ( ) 50 1 (0.02) (0.98) 50 = 0.92 =0

10 Discrete Distributions Binomial Distribution-Eample The mean number of defectives is E(X ) = np = = 1 Variance V (X ) = npq = = 0.98 Discrete Distributions Geometric Distribution PDF p() = { pq 1, = 1, 2,... Epectation E(X ) = 1 p Variance V (X ) = q p 2

11 Discrete Distributions Negative Binomial Distribution PDF p() = Epectation { ( y 1 k 1) p k q y k, y = k, k + 1, k + 2,... E(X ) = k p Variance V (X ) = kq p 2 Discrete Distributions Geometric-Negative Binomial Distribution-Eample 40% of the assembled ink-jet printers are rejected at the inspection station. Find the probability that the first acceptable ink-jet printer is the third one inspected. If we let X be the number of trials to achieve the first success (acceptable printer), p(3) = (0.4) 2 (0.6) = The probability that the third printer inspected is the second acceptable printer is ( ) 3 1 p(3) = (0.4) 3 2 (0.6) 2 =

12 Discrete Distributions Poisson Distribution PDF and CDF p() = { e α α!, = 0, 1, 2,... F (X ) = Epectation and Variance e α α i i=0 E(X ) = α V (X ) = α i! Discrete Distributions Poisson Distribution-cont. p() F() (a) (b) Figure : Poisson PMF and CDF

13 Discrete Distributions Poisson Distribution-Eample A computer repairman is beeped each time there is a call for service. The number of beeps per hour is a Poisson RV with a mean of α = 2 per hour. The probability of 3 beeps in the net hour is given by p(3) = e 2 (2) 3 3! = 0.18 = F (3) F (2) = The probability of 2 or more beeps in a 1-hour period is P(X 2) = 1 P(X < 2) = 1 F (1) = Uniform Distribution PDF and CDF f () = F (X ) = Epectation and Variance { 1 b a, a b 0, < a a < b 1, b a b a, E(X ) = a + b 2 V (X ) = (b a)2 12

14 Uniform Distribution-cont. f() F() Figure : Uniform PDF and CDF Eponential Distribution PDF and CDF { λe f () = λ, 0 { 1 e F (X ) = λ, 0 Epectation and Variance E(X ) = 1 λ V (X ) = 1 λ 2

15 Eponential Distribution-cont. One of the most important properties of the eponential distribution is that it is memoryless, which means, for all s 0 and t 0, P(X > s + t X > s) = P(X > t) = P(X > s + t) P(X > s) = e λ(s+t) e λs = e λt = P(X > t) Eponential Distribution-cont. f() F() Figure : Uniform PDF and CDF

16 Eponential Distribution-cont. f() Figure : PDFs for Several Eponentials Eponential Distribution-Eample The life time of a lamp, in thousand hours, is eponentially distributed with failure rate λ = 1/3. Find the probability that the lamp will last for another 1,000 hours given that it is working after 2,500 hours. P(X > 3.5 X > 2.5) = P(X > 1) = e 1/3 = 0.717

17 Gamma Distribution PDF and CDF F (X ) = f () = { { βθ Γ(β) (βθ)β 1 e βθ, > 0 1 Epectation and Variance βθ Γ(β) (βθt)β 1 e βθt dt, > 0 E(X ) = 1 θ V (X ) = 1 βθ 2 Gamma Distribution-cont. Gamma Function Γ(β) = 0 β 1 e d We can show that Γ(β) = (β 1)! When β = k, the distribution is called as the Erlang distribution of order k.

18 Gamma Distribution-cont. f() β = 3 β = β = Figure : PDFs for Several Gammas Normal Distribution PDF and CDF [ 1 f () = ep ], ( µ)2 < < 2πσ 2 2σ 2 F (X ) = [ 1 ep ]dt, (t µ)2 < < 2πσ 2 2σ 2 It is difficult to compute the above integral analytically. We can do it numerically, but it would be dependent on the mean and the variance. Not to be so, we can transform it to standard normal distribution.

19 Normal Distribution-cont. Let z = (t µ)/σ. Then, F (X ) = P(X ) = P = ( Z µ ) σ ( µ)/σ ( µ)/σ 1 2π ep = φ(z)dz ( ) µ = Φ σ ) ( z2 dz 2 Normal Distribution-cont. Hence, X N(µ, σ 2 ) and Z N(0, 1). The PDF of Z is ( ) φ(z) = 1 2π ep, < < z2 2 The CDF is computed and tabulated for use. z ( ) 1 Φ(z) = ep t2 dt 2π 2

20 Normal Distribution-cont. f() m Figure : PDF for Normal Normal Distribution-cont. f(z) s 2 1 m 0 z Figure : PDF for Standard Normal

21 Normal Distribution-cont. f() s 2 9 m 50 (a) 0 56 f(z) s 2 1 m 0 (b) 2 z Figure : from Normal to Standard Normal Normal Distribution-Eample X is approimated by a normal distribution with mean 25 and variance 9. Compute the value for X that will be eceeded only 5% of the time. ( P(X > 0 ) = P Z > ) ( ) = 1 Φ 3 3 Hence, ( ) ( ) Φ = 0.05 Φ = Φ(1.645) = = =

22 Normal Distribution-Eample f () s m 25 (a) 0 f(z) s m 0 (b) z.05 z Figure : Normal Distribution Eample Weibull Distribution-cont. PDF and CDF f () = { β α F (X ) = ( v ) [ β 1 α ep { [ 1 ep ) β ], v ( v α ) β ], v ( v α

23 Weibull Distribution-cont. f () b 1 2 b b 1 b Figure : Several PDFs for Weibull Weibull Distribution-cont. Epectation and Variance E(X ) = v + αγ ( ) 2 V (X ) = α (Γ 2 β + 1 ( ) 1 β + 1 [ Γ ( )] ) 1 2 β + 1

24 Triangular Distribution PDF and CDF 2( a) (b a)(c a), a b f () = 2(c ) (c b)(c a), b < c 0, a ( a) 2 (b a)(c a) F (X ) =, a < b 1 (c )2 (c b)(c a), b < c 1, > c Triangular Distribution-cont. f () Height 2/(c a) a b c Figure : PDF for Triangular

25 Triangular Distribution-cont. f () 0.2 a 0 b 2 Mode 3.7 Median 4 Mean c 10 Figure : Mode, Median and Mean Lognormal Distribution PDF f () = Epectation and Variance 1 (ln µ)2 2πσ ep [ ], 0 2σ 2 2(c ) (c b)(c a), b < c E(X ) = e µ+σ2 /2 V (X ) = e 2µ+σ2 (e σ2 1)

26 Lognormal Distribution-cont f() Figure : PDF for Lognormal Beta Distribution where PDF f () = { β 1 1 (1 ) β 2 1 B(β 1,β 2 ), 0 < < 1 B(β 1, β 2 ) = Γ(β 1 )Γ(β 2 )/Γ(β 1 + β 2 )

27 Beta Distribution-cont f () Figure : Several PDFs for Beta Poisson Process Poisson Process The counting process {N(t), t 0} is said to be a Poisson process with mean rate λ if arrivals occur one at a time {N(t), t 0} has stationary increments {N(t), t 0} has independent increments We can show that P[N(t) = n] = e λt (λt) n n! for t 0 and n = 0, 1, 2,... which is a Poisson distribution with parameter α = λt. Hence, the mean and variance are both α = λt. We can show that the inter-arrival times are eponential.

28 Empirical Distributions Empirical Distributions An empirical distribution might be discrete or continuous. Its parameters are the observed values in a sample dataset. We might use an empirical distribution when it is not possible or not necessary to establish an RV has a particular parametric distribution. One advantage of an empirical distribution is that no assumption is needed beyond the observed values, but it is also a disadvantage because the sample might not cover the entire range of possible values. Summary Summary Reading HW: Chapter 5. Chapter 5 Eercises.