Scribd
Upload
77 views8 pages

CAD Algorithms: Simulated Annealing (SA)

Simulated annealing is an algorithm for finding optimal solutions to problems by imitating the physical process of annealing in solids. It allows for non-optimal moves to be accepted with a probability that decreases over time, avoiding getting stuck in local minima. The algorithm starts with a random solution and temperature and iteratively modifies the solution, accepting changes that decrease cost and sometimes those that increase it, while slowly decreasing the temperature. This process continues until an end condition is reached.

Uploaded by

chilledkarthik
Copyright
© Attribution Non-Commercial (BY-NC)
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
77 views8 pages

CAD Algorithms: Simulated Annealing (SA)

Simulated annealing is an algorithm for finding optimal solutions to problems by imitating the physical process of annealing in solids. It allows for non-optimal moves to be accepted with a probability that decreases over time, avoiding getting stuck in local minima. The algorithm starts with a random solution and temperature and iteratively modifies the solution, accepting changes that decrease cost and sometimes those that increase it, while slowly decreasing the temperature. This process continues until an end condition is reached.

Uploaded by

chilledkarthik
Copyright
© Attribution Non-Commercial (BY-NC)
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/ 8

CAD Algorithms

Simulated Annealing (SA)


Mohammad Tehranipoor
ECE Department

16 September 2008

Overview
Assume that the simulation studies a collection of M helium atoms in a cube. The position of each atom is described by three parameters that give its coordinates within the cube. The energy of this system is given by the sum of all pair-wise interaction energies. To calculate the average energy of this system, a simple Monte Carlo simulation should not be used.
A method that estimates possible outcomes from a set of random variables by simulating a process a large number of times and observing the outcomes.

This is because a random placement of the M atoms may place two of the atoms so close together that their interaction energy is virtually infinite. This adds an infinite energy to the ensemble of atom distributions and produces an infinite average energy.
16 September 2008 2

Cont.
In the real world, two helium atoms would never get that close together.
Thus, such placement will be considered a solution.

A modification to the simple Monte Carlo simulation needs to be made so that unrealistic samples are not placed into the ensemble. Such a modification was proposed in 1953 by Nicholas Metropolis and coworkers. This modified procedure is known as a Metropolis Monte Carlo Simulation.
16 September 2008 3

Metropolis Monte Carlo Simulation


In contrast with the simple Monte Carlo simulation, a new point in search space is sampled by making a slight change to the current point.
For example, a new orientation of the helium atoms is created by making a random, small change to each atom's coordinates.

If the energy of this new orientation is less than that of the old, this orientation is added to the ensemble. If the energy rises, a Boltzmann acceptance criteria is used.
If the energy rise is small enough, the new orientation is added to the ensemble. If the energy rise is too large, the new orientation is rejected and the old orientation is again added to the ensemble.

16 September 2008

Simulated Annealing
In 1983, Kirkpatrick and coworkers proposed a method of using a Metropolis Monte Carlo simulation to find the lowest energy (most stable) orientation of a system. The method is based upon the procedure used to make the strongest possible glass. This procedure heats the glass to a high temperature so that the glass is a liquid and the atoms can move relatively freely.

16 September 2008

Simulated Annealing
The temperature of the glass is slowly lowered so that at each temperature the atoms can move enough to begin adopting the most stable orientation. If the glass is cooled slowly enough, the atoms are able to "relax" into the most stable orientation. This slow cooling process is known as annealing, and so their method is known as Simulated Annealing.

16 September 2008

SA (Cont.)
Simulated Annealing (SA) is commonly said to be the oldest among the metaheuristics and surely one of the first algorithms that had an explicit strategy to avoid local minima. The fundamental idea is to allow moves resulting in solutions of worse quality than the current solution (uphill moves) in order to escape from local minima. The probability of doing such a move is decreased during the search.

16 September 2008

SA Basic Steps
The algorithm starts by generating an initial solution (either randomly or heuristically constructed) and by initializing the so-called temperature parameter T. The following is repeated until the termination condition is satisfied:
A solution s1 from the neighborhood N(s) of the solution s is randomly sampled and it is accepted as new current solution depending on f(s), f(s1) and T. s1 replaces s if f(s1) < f(s) or, in case f(s1) >= f(s), with a probability which is a function of T and f(s1) - f(s). The probability is generally computed following the Boltzmann distribution e-(f(s1) - f(s))/T.

16 September 2008

SA Basic Steps (Cont.)


The temperature T is decreased during the search process, thus at the beginning of the search the probability of accepting uphill moves is high and it gradually decreases, converging to a simple iterative improvement algorithm. This process is similar to the annealing process of metals and glass, which assume a low energy configuration when cooled with an appropriate cooling schedule. The search process shows that the algorithm is the result of two combined strategies: random walk and iterative improvement.
16 September 2008 9

SA
In the first phase of the search, the bias toward improvements is low and it permits the exploration of the search space; this erratic component is slowly decreased thus leading the search to converge to a (local) minimum. The probability of accepting uphill moves is controlled by two factors:
The difference of the objective functions The temperature.

At fixed temperature, the higher the difference f(s1)- f(s), the lower the probability to accept a move from s to s1. The higher T, the higher the probability of uphill moves.

16 September 2008

10

Basic Simulated Annealing


s = Generate_Initial_Solution() T = T_0 WHILE termination conditions not met
s1 = Pick_At_Random(N(s)) IF f(s1) < f(s) s = s1 ELSE Accept s1 as new solution with probability p(T,s1,s) ENDIF

Update(T) ENDWHILE

16 September 2008

11

SA Algorithm: With More Details


Select an initial temperature t_0 (a large number) Select an initial solution s Select a cost function f Select a neighborhood structure for the solution space Select a temperature reduction function alpha( t ) Repeat
Repeat
randomly select s1 in N(s) diff = f(s1) - f(s) if diff< 0 then s = s1 else generate random x in (0, 1) and if x < e -diff/t then s = s1

until iteration count = max_number_iteration t = alpha(t)

until stopping condition s is the approximation solution


16 September 2008 12

Simulated Annealing Flow Chart

16 September 2008

13

Advantages
The main problem with Hill Climbing is getting stuck in a local minimum. Simulated annealing works as a hill-climbing search method that allows moves in less good goal directions once in a while to escape local minima. Simulated annealing is proven to converge to the optimum solution of a problem.

16 September 2008

14

Disadvantages
Although it is proven to converge to the optimum, it converges in infinite time. Not only for this reason, but also since you have to cool down slowly, the algorithm is usually not faster than its cotemporaries.

16 September 2008

15

Cooling Schedules
Various functions can be used: alpha(t) = a*t with a between 0.8 and 0.99 alpha(t) = t / (1 + b*t), with b small (around 0)

16 September 2008

16

You might also like