Transcript Document
Stochastic Optimization
Review of Probability Theory
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Objectives
To introduce the concept of probability
To define random variables and its statistical properties
To introduce commonly used probability distributions
2
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Introduction
Most water resources decision problems face the risk of uncertainty
Uncertainty - Randomness of the variables
Hydrologic random variables: rainfall in a command area, inflow to a
reservoir, evapo-transpiration of crops etc.
Optimization models developed for water resources management - optimal
decisions with an indication of the associated hydrologic uncertainty
Two classical approaches to deal with the hydrologic uncertainty in
optimization models are:
• Implicit Stochastic Optimization (ISO)
• Explicit Stochastic Optimization (ESO)
3
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Introduction…
Implicit Stochastic Optimization (ISO)
Hydrologic uncertainty is implicitly incorporated
Optimization model is a deterministic model
Hydrologic inputs are varied with a number of equi-probable sequences
Deterministic optimization model is run once with each of the input sequences
Output set is then statistically analyzed to generate a set of optimal decisions.
4
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Introduction…
Explicit Stochastic Optimization (ESO)
Stochastic nature of the inputs is explicitly included through their probability
distributions
Optimization model is a stochastic model
A single run of the model specifies the optimal decisions
Two commonly used ESO techniques are:
Chance Constrained Linear Programming (CCLP) and
Stochastic Dynamic Programming (SDP) (will be discussed in the following lectures)
A background of probability theory is essential for ESO
5
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Concept of Probability
Sample space S: Area containing all possible outcomes of an experiment
An event is one subset of these outcomes
Probability is a measure of the likelihood of occurrence of an event
Probability can be assessed in two ways:
1.
Objective or posterior probability which is based on the observation of events and
2.
Subjective or prior probability which is based on experience or judgement.
Three basic axioms of probability are:
(i)
Totality:
P(S) = 1 where S is the sample space
(ii)
Nonnegativity: P(A) ≥ 0 where A is an event
(iii) Mutually exclusive: If A and B are two mutually exclusive events, then
P A B = P(A) + P(B)
For mutual exclusive events P A B = 0.
The third axiom after relaxing mutual exclusiveness will be P = P(A) + P(B) - P A B
6
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Random Variable
Random variable (r.v): Variable whose value is not known or cannot be measured
with certainty (or is nondeterministic)
Examples of random variables in water resources: Rainfall, streamflow, time
between hydrologic events (e.g. floods of a given magnitude), evaporation from a
reservoir, groundwater levels, re-aeration and de-oxygenation rates etc.
Any function of a random variable is also a random variable
Random variable is denoted using an upper case letter and the corresponding lower
case letter is used to denote the value that it takes
For example, daily rainfall may be denoted as X. The value it takes on a particular
day is denoted as x.
We then associate probabilities with events such as X ≥ x, 0 X x
7
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Random Variable…
Random variable can be essentially classified into two categories:
Discrete
Continuous.
Discrete r.v.:
X can take on only discrete values x1, x2, x3, ...,.
Eg.: Number of rainy days in a year which may take values such as, 10, 20, etc.
Can assume a finite number of values
Continuous r.v.:
X can take on all real values in a range
Most variables in hydrology are continuous random variables
Number of values that a continuous random variable can assume is infinite.
8
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Probability Distributions
For
discrete
random
variables,
the
probability
distribution
is
called
a
probability mass function
For
continuous random variables, the probability distribution is called a
probability density function (pdf)
The cumulative distribution function (CDF), F(x), represents the probability that X is
less than or equal to x,
i.e.
F(x) = P(X x).
The probability mass function (PMF) of X is defined as
p(x) = P(X = x).
9
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Probability Distributions…
(a) PMF and (b) CDF of a discrete random variable
F(x)
p(x)
1
. . .
F(x2)
. . .
x 1 x2 x3
. . . xN-1 x
N
x
(a)
F(x2) = p(x1) + p(x2)
x1 x2 x 3
. . . xN-1 x
N
x
(b)
For a discrete random variable, there are spikes of probability associated with the
values that the random variable assumes.
CDF appears as a staircase
10
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Probability Distributions…
For a continuous random variable, the probability density function (PDF) is:
f x
dF x
dx
where F(x) is the CDF of X.
(a) PDF and (b) CDF of a continuous random variable
F(x)
x2
F x f x dx
2
f(x)
1
F(x2)
0
x2
(a)
11
Water Resources Planning and Management: M6L1
x
x2
x
(b)
D Nagesh Kumar, IISc
Probability Distributions…
Probability distributions of continuous random variables are smooth curves
CDF of a continuous random variable denoted by F(x) is a non-decreasing function
with a maximum value of 1
CDF represents the probability that X is less than or equal to x,
i.e. F(x) = P(X x).
Any function f(x) defined on the real line can be a valid probability density function
if and only if
i.
f(x) 0 for all x, and
ii.
f(x) 1 for all x.
Given the PMF or PDF, the CDF can be obtained as
F x
p x
i i n
F x
for discrete random variables
x
f x dx
12
i
Water Resources Planning and Management: M6L1
for continuous random variables
D Nagesh Kumar, IISc
Probability Distributions…
Area under the curve to the left of x = a is
f(x)
Area Pa X b
P(X ≤ a)
Area under the curve to the left of x = b is
P(X ≤ b)
Area between x = a and x = b is
P[a ≤ X ≤ b].
0
a
b
x
Probability density function
For a continuous random variable, probability of the random variable taking a value
exactly equal to a given value is zero because
d
P X d Pd X d f x dx 0
d
13
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Probability Distributions…
F(x)
0.8
F-1(0.5) = 10
F-1(0.8) = 15
0.5
10
15
x
Cumulative density function
For any given probability α, 0 ≤ α ≤ 1, the value x of the random variable can be
determined from the CDF as
x = F-1(α).
14
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Statistical properties of random variables
Population: Set of all the values taken by a random process
Sample: Subset of the population
Expected value of (X – x0)r is the rth moment of a random variable X about any
reference point X = x0.
Mathematically,
x x f xdx
E X x0
r
r
0
for continuous case
N
E X x0 xi x0 p xi
r
r
i 1
for discrete case
where E[ ] is a statistical expectation operator.
The first three moments describe the central tendency, variability and asymmetry
of the distribution of a random variable.
15
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Statistical properties of random variables…
Expected Value or Mean
The central tendency is expressed as an expectation as
EX
x f x dx
for continuous case
N
E X xi p xi
for discrete case
i 1
The mean of a r.v is denoted by μ is equal to the expected value,
i.e., μ = E[X].
16
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Statistical properties of random variables…
Variance
Second order central moment.
Variance of a continuous r.v. is defined as
Var X 2 E X
2
x f x dx
2
The positive square root of variance is called the standard deviation, σ.
Coefficient of variation
C v
Skewness
The asymmetry of PDF of a r.v. is measured by skew coefficient
Defined as
17
E X 3 3
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Example
Probability density function (pdf) of a random variable X is
f(x) = 6 x2
= 0
0 x1
else where
Determine (1) Cumulative distribution function (cdf); (2) Expected value, E(X); (3)
Variance, Var (X); (4) P[X 0.6]; and (5) P[0.4 X 0.7]
Solution:
1.
Cumulative distribution function
x
F x f x dx
x
6 x 2 dx 2 x 3
0
18
Water Resources Planning and Management: M6L1
0 x 1
D Nagesh Kumar, IISc
Example…
2. Expected value, E(X):
3. Variance, Var (X)
1
E X x f x dx x 6 x 2 dx 3 / 2
0
Var X
2
x
f x dx
1
x 3 / 2 6 x 2 dx 1.2
2
0
4. P[X 0.6]
PX 0.6 1 PX 0.6 1 F 0.6
1 2 0.63 0.568
5. P[0.4 X 0.7]
P0.4 X 0.7 PX 0.7 PX 0.4
F 0.7 F 0.4
19
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Commonly used Probability Distributions
Three commonly used distributions in water resources are: Normal, Lognormal and
Exponential distributions.
Normal distribution
Also called as Gaussian distribution
Two parameters are involved in this distribution: mean and variance
A normal random variable with mean μ and variance σ2 is denoted as
X ~ N(μ, σ2)
PDF of the normal distribution given by f(x) is expressed as
1 x 2
1
f x
exp
2
2
20
Water Resources Planning and Management: M6L1
for x
D Nagesh Kumar, IISc
Normal distribution
f(x)
PDF: Bell-shaped and symmetric at x = μ
μ
CDF of a normal distribution is
f x
x
x
Normal PDF
1 x 2
1
exp
dx
2
2
for x
Normal random variables are usually transformed to standardized variate Z with zero
mean and unit variance
i.e.,
Z = (X - μ) / σ.
Then PDF of Z can be expressed as
z2
z
exp
2
2
1
for z
Values of (z) obtained by numerical integration are used in the computations
21
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Example
The monthly streamflow at a reservoir site is represented by a random variable X which
follows normal distribution with a mean of 100 units and a standard deviation of 50
units. Find (1) P[X > 150]; (2) P[X ≤ 40] and (3) The flow value which will be
exceeded with a probability of 0.8.
Solution:
The monthly streamflow at a reservoir site is represented by a random variable X which
(1) P[X > 150]
PX 150 P X / 150 100 / 50
PZ 1 1 PZ 1
1 0.8413 0.1587
22
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Example…
(2) P[X ≤ 40]
PX 40 P X / 40 100 / 50
PZ 1.2
0.1539
(3) To find P[X ≥ x] = 0.8
P X x 0.8
PZ x / 0.8
1 PZ z 0.8
PZ z 0.2
z x 100 / 50 0.84
x 58 units
23
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc
Lognormal distribution
Used when random variable cannot be negative
A r.v. X is lognormally distributed if its logarithmic transform
Y=ln(X)
has a normal distribution with mean μlnX and variance σ2lnX
The PDF of lognormal r.v. is
1
f x
2 X ln X
24
1 ln X ln X
exp
2
ln X
Water Resources Planning and Management: M6L1
2
for X
D Nagesh Kumar, IISc
Exponential distribution
PDF of an exponential distribution with parameter λ is:
e x
f x
0
x0
x0
λ > 0 is the parameter of the distribution
Mean E[X] = 1 / λ
Var (X) = E[X2] - E[X] 2 = 1 / λ2.
CDF is given by:
1 e x
F x f x dx
0
x
25
Water Resources Planning and Management: M6L1
x0
x0
D Nagesh Kumar, IISc
Thank You
Water Resources Planning and Management: M6L1
D Nagesh Kumar, IISc