Simulated Annealing
a metaheuristic approach
Presented by;
Nitesh Bansal (2k15/the/09)
Nirmal Pratap Singh
(2k15/the/08)
1
Outline
Introduction
Basic
philosophy
Procedure : Simulated Annealing
Example : Travelling Salesman Problem
Hill Climbing
Stimulated Annealing vs. Hill Climbing
Features
Drawback
Applications
References
Introduction
Studied to obtain an optimal solution for OR models
including
Linear Programming
Non-Linear Programming
These algorithm are invaluable, but doesnt
always work.
Still it is important to find a good feasible solution
3
Basic Philosophy
Motivated by physical annealing process
Annealing
Process
Raising the temperature up to a very high level
(melting temperature, for example), the atoms have a
higher energy state and a high possibility to rearrange the crystalline structure.
Cooling down slowly, the atoms have a lower and
lower energy state and a smaller and smaller
possibility to re-arrange the crystalline structure.
cool it down according to a specific Temperature
schedule.
Ground state is a minimum energy state of the solid
4
Analogy : Simulated Annealing
Metal Problem
Energy State Cost Function
Temperature Control Parameter
A completely ordered crystalline structure
the optimal solution for the problem
System state feasible solution
Global optimal solution can be achieved as long
as the cooling process is slow enough
6
Other Names : Simulated Annealing
Monte
Carlo Annealing
Statistical Cooling
Probabilistic Hill Climbing
Stochastic Relaxation
Probabilistic Exchange Algorithm
Iterative Improvement in SA
Procedure
Step1. Start with initial configuration
Step2. Repeatedly search neighborhood and select a
neighbor as candidate (by Sub-Tour Reversal)
Step3. Evaluate some cost function (or fitness function)
and accept candidate if "better"; if not, select
another neighbor (by Move Selection Rule)
Step4. Stop if quality is sufficiently high, if no improvement
can be found or after some fixed time.
8
Sub-Tour Reversal
Adjusts
the sequence of cities visited in the
current trial solution by selecting a
subsequence of the cities and simply
reversing the order in which that
subsequence of cities is visited.
example: 1-2-3-4-5-6-7-1
distance=69
If we select the subsequence 3-4 and reverse
1-2-4-3-5-6-7-1
distance=65
9
Temperature Schedule
Suppose five iterations are performed at
each of five values of T {T1,T2,T3,T4,T5}
Where,
T1=0.2ZC T2=0.5T1 T3=0.5T2
T4=0.5T3 T5=0.5T4
This
particular temperature schedule is
only illustrative.
10
Example of Simulated Annealing
Traveling Salesman Problem (TSP)
Given 7 cities and the traveling cost/distance
between any two cities
A salesman need to start from city 1 and travel
all other cities then back to city 1
Minimize the total traveling cost/distance
Travelling salesman problem
with seven cities
12
Step 1. Initial trial solution
may enter any feasible solution
(sequence of the cities on the tour)
lets take 1-2-3-4-5-6-7-1
Zc = 12+8+11+11+6+9+12 = 69
Step 2. Neighborhood Structure
immediate neighbour of the current trial
solution is one that is reached by making a
sub-tour reversal
13ini
Step 2.
Selecting
a sub-tour to be reversed requires
selecting the slot in the current sequence of cities
where the sub-tour currently begins and ends. (1-23-4-5-6-7-1)
The beginning slot can be anywhere except the first
and the last slot(reserved for the home city) and next
to last slot
The ending slot must be somewhere after beginning
slot, excluding the last slot
Random no. are used to give equal probabilities to
select any of the eligible beginning slots and ending
slots
14
How the random selection of an immediate
neighbour is made
15
Step 3. Evaluation of cost function and
Acceptance of a candidate
Move selection rule : To Accept or not to Accept
Objective is Minimization
ZN = obj. function value for the current trial solution
ZC = obj. function value for the current candidate to be
the next trial solution
If ZN ZC always accept this candidate
If ZN ZC accept the candidate with the following probability
P(acceptance) = ex where x= (ZC ZN) / T
(If objective is maximization instead, reverse Z N and ZC
in the above formula)
16
To Accept
The new candidate for feasible solution is
1-2-4-3-5-6-7-1 and ZN = 65
Since ZN = 65 < ZC
This candidate is better than the current trial solution, so
the candidate is automatically accepted to be the next trial
solution.
17
Not to Accept
Now considering a case where the current solution is
worse than the current trial solution.
Suppose second iteration results in reversing 3-5-6 i.e 12-4-6-5-3-7-1 ; Zc = 64
Then suppose the third iteration result in reversing 3-7 i.e
1-2-4-6-5-7-3-1 ; Zn = 66
Since objective here is minimization, the probability of
accepting 1-2-4-6-5-7-3-1 as the next trial solution isP(acceptance) = ex ,
where x= (ZC ZN) / T1 and T1 = 13.8
P = 0.865
If the next random number generated is more than 0.865
this candidate solution will be rejected.
18
To accept or not to accept :
some facts
The probability of accepting a worse state is a
function of both the temperature of the system and
the change in the cost function
As the temperature decreases, the probability of
accepting worse moves decreases
If t=0, no worse moves are accepted (i.e. hill
climbing)
19
Iterations
20
Travelling Salesman : Literally?
There
are number of applications of travelling
salesman problems that have nothing to do
with the salesman.
Example:
When a truck leaves a distribution
center to deliver goods to a number of
locations the problem of determining the
shortest route for doing this is a travelling
salesman problem.
21
Hill climbing :
A local improvement procedure
Candidate is always and
only accepted if cost is
lower (or fitness is
higher) than current
configuration
Stop when no neighbour
with lower cost (higher
fitness) can be found
Disadvantages:
Procedure converges to
one Local optimum and
then stop
22
How SA has ability to escape from
local optimum? (Minimization)
AT INIT_TEMP
Unconditional Acceptance
COST FUNCTION, C
HILL CLIMBING
Move accepted with
probability
= e-(^C/temp)
HILL CLIMBING
HILL CLIMBING
AT FINAL_TEMP
NUMBER OF ITERATIONS
23
Simulated Annealing
(Minimization)
Move from A to A accepted
if cost(A) cost(A)
or with probability
P(A,A) = e (cost(A) cost(A))/T
Flow Chart
25
Simulated Annealing
Random
function
value
Starting Point
T = Very
High
Simulated Annealing
Random
function
value
Step
T = Very
High
Simulated Annealing
Even
though E is lower, accept
function
value
T = Very
High
Simulated Annealing
Next
Step; accept since higher E
function
value
T = Very
High
Simulated Annealing
Next
Step; accept since higher E
function
value
T = Very
High
Simulated Annealing
Next
T = High
Step; accept even though lower
function
value
Simulated Annealing
Next
T = High
Step; accept even though lower
function
value
Simulated Annealing
Next
Step; accept since higher
function
value
T = Medium
Simulated Annealing
Next
T = Medium
Step; lower, but reject (T is falling)
function
value
Simulated Annealing
Next
T = Medium
Step; Accept since E is higher
function
value
Simulated Annealing
Next
T = Low
Step; Accept since E change small
function
value
Simulated Annealing
Next
Step; Accept since E larget
function
value
T = Low
Simulated Annealing
Next
T = Low
Step; Reject since E lower and T low
function
value
Simulated Annealing
Eventually
function
value
T = Low
converge to Maximum
Applications
Travelling
salesman problem
Locomotive scheduling
Optimal placement of > 106 transistors on a
chip
DNA Mapping
Financial Analysis
Banking Industry
40
Features
Ability
to escape from a local optima.
Deals
with problems that are too large and
complicated to be solved by exact algorithms.
41
Drawback
No
guarantee that the best solution found will
be an optimal solution or even a nearly
optimal solution.
Another
drawback is time consumption.
42
References
Introductions to Operation Research : concepts
and cases(TMH Publications)
- Frederick S.Hiller, Gerald J.Lieberman
Progress in Operations Research
edited by-Russell L.Ackoff
43
Thank You!
44
