Scribd
Upload
23 views14 pages

Lecture 31

Uploaded by

appmerchant22
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
23 views14 pages

Lecture 31

Uploaded by

appmerchant22
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 14

INDIAN INSTITUTE OF SCIENCE

Water Resources Systems:


Modeling Techniques and Analysis
Lecture - 31
Course Instructor : Prof. P. P. MUJUMDAR
Department of Civil Engg., IISc.

1
Summary of the previous lecture
• Deterministic equivalent of CCLP
Min K
s.t.  Dt  bt  bt 1   FQ 1  1 
1
t

R t
max

 bt  bt 1  FQt 1  2 
Min K
bt 1  K vt s.t. P  Rt  Dt   1
bt 1  S min P  Rt  Rtmax    2

P  St  K    3
bt  0
P  St  Smin    4
K 0

• Examples on writing the deterministic equivalent


of CCLP
2
STOCHASTIC DYNAMIC
PROGRAMMING
(SDP)

3
Stochastic Dynamic Programming
Evaporation
Inflow

Storage Release

• Reservoir storage and inflow divided into discrete


class intervals
..
..
.
3
2
1
• Inflow - Stochastic process.
4
Stochastic Dynamic Programming
• Stage: Time period ( e.g., month, week etc.) at which
decisions need to be taken.
• State variables: storage at the beginning of period t
and inflow during period t.
• Decision: Release from the reservoir during period t.
• Objective: To optimize (maximize or minimize) the
expected value of a system performance measure.
For example,
• Maximize the expected value of annual hydropower
generated
• Minimize the expected value of annual flood damage.
5
Stochastic Dynamic Programming
Evaporation
Inflow

Storage Release

• System performance measure is a function of


release and/or storage
• Hydropower generated
• Crop yield achieved
• Monitory benefits
• Magnitude of flood mitigation etc.
6
Stochastic Dynamic Programming
Notation: t t+1
k i l j Class intervals
Skt Qit Sl t+1 Qjt+1 Representative
values
Q : Inflow S : Storage R : Release

k : Class interval of storage at the beginning of period t


i : Class interval of inflow during period t
l : Class interval of storage at the beginning of period t+1
j : Class interval of inflow during period t+1
7
Stochastic Dynamic Programming
t t+1

k l Storage class interval

i j Inflow class interval

Storage representative
Skt Sl t+1
values

Qit Qjt+1 Inflow representative


values
All variables are discretised into a no. of class intervals
8
Stochastic Dynamic Programming
• First order Markov chain (or single step Markov
chain):
P  X t X t 1 , X t  2 ,..... X 0   P  X t X t 1 

Xt : Random variable
• Inflows during time intervals ranging from 10 days
to a year may be assumed to follow a single step
Markov chain.
• Transition probabilities are used to measure the
dependence of the inflow during period t+1 on the
inflow during the period t.

9
Stochastic Dynamic Programming
t
• The transition probability Pij is defined as the
probability that the inflow during period t+1 will be in
class interval j, given that the inflow during the
period t lies in the class interval i,
Pijt  P Qt 1  j Qt  i 
where Qt = i indicates that the inflow during the
period t belongs to the discrete class interval i.
• The transition probabilities are estimated from
historical inflow data.

10
Stochastic Dynamic Programming

• Release during the period t is Rkilt

• Storage state changes because of inflow, release


and evaporation.

• Inflow state changes randomly from period to


period.

11
Stochastic Dynamic Programming

State transformation:

Sl t 1  S k t  Qk t  Ekl t  Rkil t

Eklt is evaporation loss during period t


corresponding to storage class intervals k and l.
• For given k, i and l, release is computed as
Rkil t  Sk t  Qk t  Ekl t  Sl t 1

12
Stochastic Dynamic Programming
• System performance measure: Bkilt
• Bkilt is, in general, a function of Skt, Rkilt and Slt+1
For example,
• Amount of power generated during period t.
• Deficit release from target in period t.

13
Stochastic Dynamic Programming
Backward recursion:
Year N Year N – 1 Year 1
n = NT n=T n=T–1 n=1
t=1t=2 t=Tt=1t=2 t=T t=1t=2 t=T

Progress of computations

ft n  k , i  : Optimum expected value of the system


performance measure at stage n,
corresponding to time period is t.

14

You might also like