5.

Simulated Annealing
5.2 Advanced Concepts

Fall 2010
Instructor: Dr. Masoud Yaghini
 Simulated Annealing: Part 2
Outline

Acceptance Function

Initial Temperature

Equilibrium State

Cooling Schedule

Stopping Condition

Handling Constraints

An Exercise

References
Acceptance Function
 Simulated Annealing: Part 2
Acceptance Function

SA can escape from local optima due to the

probabilistic acceptance of a nonimproving
neighbor.

The probability of accepting is dependent of:
– The temperature T
– The change of the objective function DE
 Simulated Annealing: Part 2
Acceptance Function

The acceptance probability of a nonimproving move

is:
− ∆E
T
P(∆E , T ) = e >R
– where E is the change in the evaluation function,
– T is the current temperature, and
– R is a uniform random number between 0 and 1.
 Simulated Annealing: Part 2
To accept or not to accept?

Change Temperature DE/T)

exp(-D
0.2 0.95 0.8101
0.4 0.95 0.6563
0.6 0.95 0.5317
0.8 0.95 0.4308
0.2 0.10 0.1353
0.4 0.10 0.0183
0.6 0.10 0.0024
0.8 0.10 0.0003
 Simulated Annealing: Part 2
Acceptance Function

At high temperatures,
– the probability of accepting worse moves is high.
– If T = ∞, all moves are accepted
– It corresponds to a random local walk in the landscape.

At low temperatures,
– The probability of accepting worse moves decreases.
– If T = 0, no worse moves are accepted
– The search is equivalent to local search.

Moreover, the probability of accepting a large
deterioration in solution quality decreases
exponentially toward 0.
Initial Temperature
 Simulated Annealing: Part 2
Initial Temperature

If the starting temperature is very high,

– the search will be a random local search for a period of
time
– accepting all neighbors during the initial phase of the
algorithm.
– The main drawback of this strategy is its high
computational cost.

If the initial temperature is very low,
– the search will be a local search algorithm.

Temperature must be high enough to allow moves
to almost neighborhood state.

Problem is finding a suitable starting temperature
 Simulated Annealing: Part 2
Initial Temperature

Finding a suitable starting temperature

methods:
– Acceptance deviation method
– Tuning for initial temperature method
 Simulated Annealing: Part 2
Initial Temperature

Acceptance deviation method

– The starting temperature is computed using preliminary
experimentations by: kσ
– where

σ represents the standard deviation of difference between values
of objective functions and

k = −3/ln(p) with the acceptance probability of p, which is
greater than 3σ
 Simulated Annealing: Part 2
Initial Temperature

Tuning for initial temperature method:

– Start high, reduce quickly until about 60% of worse
moves are accepted.
– Use this as the starting temperature
Equilibrium State
 Simulated Annealing: Part 2
Equilibrium State

Once an equilibrium state is reached, the

temperature is decreased.

To reach an equilibrium state at each temperature, a
number of sufficient transitions (moves) must be
applied.

The number of iterations must be set according to:
– The size of the problem instance and
– Particularly proportional to the neighborhood size |N(s)|
 Simulated Annealing: Part 2
Equilibrium State

The strategies that can be used to determine the

number of transitions visited:
– Static strategy
– Adaptive strategy
 Simulated Annealing: Part 2
Equilibrium State

Static strategy
– The number of transitions is determined before the search
starts.
– For instance, a given proportion y of the neighborhood
N(s) is explored.
– Hence, the number of generated neighbors from a
solution s is y · |N(s)|.
– The more significant the ratio y, the higher the
computational cost and the better the results.
 Simulated Annealing: Part 2
Equilibrium State

Adaptive strategy
– The number of generated neighbors will depend on the
characteristics of the search.
– One adaptive approach is an improving neighbor solution
is generated.
– This feature may result in the reduction of the
computational time without compromising the quality of
the obtained solutions
– Another approach is achieving a predetermined number
of iterations without improvement of the best found
solution in the inner loop with the same temperature
Cooling Schedule
 Simulated Annealing: Part 2
Cooling Schedule

In the SA algorithm, the temperature is decreased

gradually such that Ti > 0, ∀i

There is always a compromise between the quality
of the obtained solutions and the speed of the
cooling schedule.

If the temperature is decreased slowly, better
solutions are obtained but with a more significant
computation time.
 Simulated Annealing: Part 2
Cooling Schedule

The temperature can be updated in different ways:

– Static Strategy

Linear
– Dynamic Strategy

Geometric

Logarithmic
– Adaptive Strategy
 Simulated Annealing: Part 2
Cooling Schedule

Linear
– In the trivial linear schedule, the temperature T is updated
as T = T − β, where β is a specified constant value.
– Hence, we have
Ti = T0 − i × β
– where Ti represents the temperature at iteration i.
– β is a specified constant value
– T0 is the initial temperature
 Simulated Annealing: Part 2
Cooling Schedule

Geometric
– In the geometric schedule, the temperature is updated
using the formula
Ti+1 = α.Ti
– where α ∈]0, 1[.
– It is the most popular cooling function.
– Experience has shown that α should be between 0.5 and
0.99.
 Simulated Annealing: Part 2
Cooling Schedule

Logarithmic
– The following formula is used:

T0
Ti =
log(i + 10)
– This schedule is too slow to be applied in practice but has
the property of the convergence proof to a global
optimum.
 Simulated Annealing: Part 2
Cooling Schedule

Adaptive Strategy
– Most of the cooling schedules are static or dynamic in the
sense that the cooling schedule is defined completely a
priori.
– In this case, the cooling schedule is “blind” to the
characteristics of the search landscape.
– In an adaptive cooling schedule, the decreasing rate is
depends on some information obtained during the search.
Stopping Condition
 Simulated Annealing: Part 2
Stopping Condition

Concerning the stopping condition, theory suggests

a final temperature equal to 0.

In practice, one can stop the search when the
probability of accepting a move is negligible.
 Simulated Annealing: Part 2
Stopping Condition

The following stopping criteria may be used:

1. Reaching a final temperature TF is the most popular
stopping criteria.
o This temperature must be low (e.g., Tmin = 0.01).
2. Achieving a predetermined number for successive
temperature values no improvement in solution quality
3. After a fixed amount of CPU time
4. When the objective reaches a pre-specified threshold
value
Handling Constraints
 Simulated Annealing: Part 2
Handling Constraints

Constraints cannot handled implicitly

– Penalty function approach should be used

Constraints
– Hard Constraints: these constraints cannot be violated
in a feasible solution
– Soft Constraints: these constraints should, ideally, not
be violated but, if they are, the solution is still feasible
 Simulated Annealing: Part 2
Handling Constraints

Hard constraints are given a large weighting.

– The solutions which violate those constraints have a high
cost function

Soft constraints are weighted depending on their
importance

Weightings can be dynamically changed as the
algorithm progresses.
– This allows hard constraints to be accepted at the start of
the algorithm but rejected later
An Exercise
 Simulated Annealing: Part 2
The Knapsack problem

There are n items:
– Each item i has a weight wi
– Each item i has a value vi

The knapsack has a limited
capacity of W units.

We can take one of each
item at most

max ∑v *x
i
i i i = 1,2,..., n

subject to ∑ wi * xi ≤ W
i
x ∈ {0,1}
i
 Simulated Annealing: Part 2
The Knapsack problem

Example:
– Item: 1 2 3 4 5 6 7
– Benefit: 5 8 3 2 7 9 4
– Weight: 7 8 4 10 4 6 4

Knapsack holds a maximum of 22 pounds

Fill it to get the maximum benefit

The problem description:

– Maximize ∑v i
i

– While
∑w
i
i ≤W
 Simulated Annealing: Part 2
Solution 1

Item 1 2 3 4 5 6 7

Solution 1 1 0 0 1 0 0
Benefit 5 8 3 2 7 9 4
Weight 7 8 4 10 4 6 4

Objective: 5 + 8 + 7 = 20

Weight: 7 + 8 + 4 = 19 <= 22
 Simulated Annealing: Part 2
Solution 2 overweighted

Item 1 2 3 4 5 6 7
Solution 0 1 0 1 0 1 0

Benefit 5 8 3 2 7 9 4

Weight 7 8 4 10 4 6 4

Weight: 8 + 10 + 6 = 24 > 22
References
 Simulated Annealing: Part 2
References

El-Ghazali Talbi, Metaheuristics : From Design to

Implementation, John Wiley & Sons, 2009.

J. Dreo A. Petrowski, P. Siarry E. Taillard, Metaheuristics
for Hard Optimization, Springer-Verlag, 2006.
The End